Corpus-Complete Solution Map
Snapshot: corpus-complete-20260711 (2026-07-11) · source commit 1737631198b184e05ba0a343d67df6fa9e71b938
Scope: 1780 approved schema-v2 extractions; 1754 deduped sources; 991 deduped agentic projects.
Corpus-complete analysis output. Counts are generated from extraction.json; citations link to deterministic report.md views. This is an input to, not the output of, Phase 4 synthesis.
For each problem class, this report shows adopted building blocks, recurring pattern combinations, compressed solution-shape signatures, and representative rejected alternatives and tradeoffs.
Denominator semantics: each class denominator is its own deduplicated any-label membership. Classes overlap, so class totals are not a partition of the 1754-source corpus.
Evaluation, observability, and reliability governance
Deduped projects: 1326. Industries: 16. Provenance: 1242 source-backed / 84 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
eval-driven-development |
763 |
57.5% |
tool-use-orchestration |
499 |
37.6% |
human-in-the-loop-gate |
490 |
37.0% |
rag-retrieval |
440 |
33.2% |
observability-tracing |
420 |
31.7% |
multi-agent-orchestration |
266 |
20.1% |
model-routing |
259 |
19.5% |
safety-policy-enforcement |
256 |
19.3% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
eval-driven-development + tool-use-orchestration |
10 |
0.8% |
eval-driven-development + human-in-the-loop-gate + rag-retrieval |
6 |
0.5% |
human-in-the-loop-gate + rag-retrieval |
6 |
0.5% |
context-engineering + eval-driven-development |
5 |
0.4% |
| none |
5 |
0.4% |
eval-driven-development + human-in-the-loop-gate + llm-as-judge |
4 |
0.3% |
eval-driven-development + multi-agent-orchestration |
4 |
0.3% |
multi-agent-orchestration + tool-use-orchestration |
4 |
0.3% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
149 |
11.2% |
adopting-model-context-protocol-mcp-in-financial-services-for-ai-system-integration (evergreen_wealth_/bloomberg/_saxo_bank, 2026)
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024)
agentic-system-for-autonomous-code-monitoring-and-maintenance (ramp, 2026) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
114 |
8.6% |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026)
agent-driven-development-for-ai-research-using-github-copilot-cli (github, 2026)
agent-driven-ui-framework-migration-at-enterprise-scale (block, 2026) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
66 |
5.0% |
ai-agent-for-automated-merchant-classification-and-transaction-matching (ramp, 2025)
ai-powered-client-services-assistant-for-post-trade-services (london_stock_exchange_group, 2025)
ai-powered-insurance-claims-chatbot-with-continuous-feedback-loop (allianz, 2023) |
| retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline |
62 |
4.7% |
ai-agent-system-for-automated-travel-itinerary-generation (aimpoint_digital, 2024)
ai-assisted-database-debugging-platform-at-scale (databricks, 2024)
ai-powered-clinical-outcome-assessment-review-using-generative-ai (clario, 2025) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
61 |
4.6% |
ai-agent-for-automated-merchant-classification-correction (ramp, 2025)
ai-powered-betting-assistant-for-sports-wagering-platform (fanduel, 2025)
ai-powered-clinical-documentation-with-multi-region-healthcare-compliance (heidi_health, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
59 |
4.4% |
accelerating-llm-inference-with-speculative-decoding-for-ai-agent-applications (linkedin, 2025)
advanced-speaker-diarization-and-attributed-transcription-using-deep-learning-models (pyannoteai, 2026)
agent-based-ai-assistants-for-enterprise-and-e-commerce-applications (prosus, 2024) |
| retrieval-grounded assistant + evaluation-governed pipeline + cost/latency optimized serving |
50 |
3.8% |
ai-powered-crm-insights-with-rag-and-text-to-sql (tp_icap, 2025)
bert-based-sequence-models-for-contextual-product-recommendations (instacart, 2024)
building-a-comprehensive-llm-platform-for-food-delivery-services (swiggy, 2024) |
| evaluation-governed pipeline + guarded enterprise assistant + context/memory system |
48 |
3.6% |
ai-powered-consent-education-tool-for-preventing-gender-based-violence (override, 2026)
ai-powered-contact-center-transformation-for-pet-retail (petco, 2024)
ai-powered-contact-center-transformation-for-student-support-services (anthology, 2024) |
Decisions and Rejected Alternatives
Extracted alternatives: 4707 (3.55 per raw project). Tradeoffs: 4233 (3.19 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Document standards in wikis and SOPs |
Written process docs depended on humans remembering and applying them, whereas skills can make the golden path easier to enforce inside the agent workflow. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Let Claude Code write generic PR descriptions |
The generic descriptions summarized code rather than intent, which was not useful enough for human or agent reviewers. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Optimize token spend early |
Intercom accepted high model spend during the adoption phase because the perceived throughput alpha was worth learning faster. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Traditional lexical and embedding metrics |
They caught extreme drift but missed citation gaps, hallucinated filenames, formatting errors, and factual failures. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
speed and learning over early token-cost optimization |
The source frames high model spend as an investment while throughput was increasing. |
summary / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
skills and hooks embedded in Claude Code workflows over wiki/SOP guidance that humans must remember |
Intercom treated skills as upstream guardrails analogous to CI/CD, making standards easier to follow before code reached review. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
a create-PR skill plus LLM judging over accepting Claude Code's default code-summary descriptions |
Default descriptions lost the intent reviewers needed, so Intercom codified its PR-description expectations and measured quality over historical data. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
Automated evaluation gates on every material pipeline change over Faster ad-hoc experimentation without hard regression thresholds |
Dropbox accepted extra evaluation overhead because conversational AI regressions could appear in any stage of the pipeline. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
Enterprise knowledge retrieval and RAG
Deduped projects: 1063. Industries: 16. Provenance: 997 source-backed / 66 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
rag-retrieval |
520 |
48.9% |
eval-driven-development |
498 |
46.8% |
tool-use-orchestration |
411 |
38.7% |
human-in-the-loop-gate |
340 |
32.0% |
observability-tracing |
272 |
25.6% |
structured-output-contract |
193 |
18.2% |
safety-policy-enforcement |
191 |
18.0% |
multi-agent-orchestration |
185 |
17.4% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
eval-driven-development + human-in-the-loop-gate + rag-retrieval |
6 |
0.6% |
eval-driven-development + tool-use-orchestration |
6 |
0.6% |
human-in-the-loop-gate + rag-retrieval |
6 |
0.6% |
multi-agent-orchestration + rag-retrieval |
5 |
0.5% |
data-quality-loop + eval-driven-development + human-in-the-loop-gate + observability-tracing + rag-retrieval |
4 |
0.4% |
eval-driven-development + fine-tuning-adaptation + retrieval-ranking |
4 |
0.4% |
rag-retrieval + safety-policy-enforcement + tool-use-orchestration |
4 |
0.4% |
rag-retrieval + tool-use-orchestration |
4 |
0.4% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline |
69 |
6.5% |
ai-agent-system-for-automated-travel-itinerary-generation (aimpoint_digital, 2024)
ai-assisted-database-debugging-platform-at-scale (databricks, 2024)
ai-powered-clinical-outcome-assessment-review-using-generative-ai (clario, 2025) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
67 |
6.3% |
ai-agent-for-automated-merchant-classification-and-transaction-matching (ramp, 2025)
ai-powered-client-services-assistant-for-post-trade-services (london_stock_exchange_group, 2025)
ai-powered-menu-description-generation-for-restaurant-platforms (doordash, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
65 |
6.1% |
adopting-model-context-protocol-mcp-in-financial-services-for-ai-system-integration (evergreen_wealth_/bloomberg/_saxo_bank, 2026)
ai-agent-development-and-evaluation-platform-for-insurance-underwriting (snorkel, 2025)
ai-agents-for-automated-product-quality-testing-and-bug-detection (coinbase, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + guarded enterprise assistant |
63 |
5.9% |
agentic-ai-for-cloud-migration-and-application-modernization-at-scale (commonwealth_bank_of_australia, 2025)
agentic-ai-systems-for-drug-discovery-and-business-intelligence (loka, 2025)
agentic-data-analyst-for-enterprise-analytics (ramp, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
59 |
5.6% |
agent-driven-development-for-ai-research-using-github-copilot-cli (github, 2026)
agent-driven-ui-framework-migration-at-enterprise-scale (block, 2026)
agentic-ai-search-with-custom-evaluation-framework-for-church-management (pushpay, 2026) |
| retrieval-grounded assistant + multi-agent/tool workflow + context/memory system |
54 |
5.1% |
agentic-ai-for-title-operations-workflow-optimization (rocket, 2026)
agentic-ai-system-for-document-summarization-and-analysis (moveworks, 2024)
agentic-ai-systems-for-legal-tax-and-compliance-workflows (thomson_reuters, 2024) |
| retrieval-grounded assistant + evaluation-governed pipeline + cost/latency optimized serving |
50 |
4.7% |
ai-powered-crm-insights-with-rag-and-text-to-sql (tp_icap, 2025)
bert-based-sequence-models-for-contextual-product-recommendations (instacart, 2024)
building-a-comprehensive-llm-platform-for-food-delivery-services (swiggy, 2024) |
| retrieval-grounded assistant + evaluation-governed pipeline + context/memory system |
44 |
4.1% |
agentic-rag-implementation-for-retail-personalization-and-customer-support (mongodb, 2024)
agentic-search-for-multi-source-legal-research-intelligence (harvey, 2025)
ai-powered-legal-document-review-and-analysis-platform (lexbe, 2024) |
Decisions and Rejected Alternatives
Extracted alternatives: 3751 (3.53 per raw project). Tradeoffs: 3352 (3.15 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Traditional lexical and embedding metrics |
They caught extreme drift but missed citation gaps, hallucinated filenames, formatting errors, and factual failures. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Rejected alternative |
Public benchmark datasets alone |
Public datasets lacked the long-tail phrasing and content mix of real Dash users. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Rejected alternative |
Spreadsheet-managed experiments |
Spreadsheets made results scattered, hard to reproduce, and hard to compare across model or prompt versions. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Rejected alternative |
Only offline or staging evaluation |
Dropbox needed asynchronous live-traffic scoring to catch silent degradation from real user queries without adding latency. |
article / implied |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
Automated evaluation gates on every material pipeline change over Faster ad-hoc experimentation without hard regression thresholds |
Dropbox accepted extra evaluation overhead because conversational AI regressions could appear in any stage of the pipeline. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
LLM judges with calibrated rubrics over BLEU, ROUGE, BERTScore, and embedding similarity alone |
Dropbox needed factuality, citation, formatting, and tone checks that traditional text metrics did not enforce. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
layered merge, stage, and production gates over one large evaluation pass late in release |
Layered gates let Dropbox block obvious regressions early and catch scale-only issues in production-like or live settings. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
A hybrid transformer-encoder/RNN-decoder distilled from Pegasus over Serving the larger Pegasus model directly in the product path |
Google traded architectural simplicity for lower latency and memory while keeping similar summarization quality. |
article / stated |
abstractive-conversation-summarization-for-google-chat-spaces (google, 2022) |
Coding-agent and software delivery automation
Deduped projects: 1022. Industries: 16. Provenance: 961 source-backed / 61 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
eval-driven-development |
568 |
55.6% |
tool-use-orchestration |
480 |
47.0% |
human-in-the-loop-gate |
400 |
39.1% |
observability-tracing |
312 |
30.5% |
safety-policy-enforcement |
217 |
21.2% |
multi-agent-orchestration |
212 |
20.7% |
rag-retrieval |
205 |
20.1% |
structured-output-contract |
195 |
19.1% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
eval-driven-development + tool-use-orchestration |
9 |
0.9% |
| none |
7 |
0.7% |
context-engineering + eval-driven-development |
4 |
0.4% |
multi-agent-orchestration + tool-use-orchestration |
4 |
0.4% |
observability-tracing + tool-use-orchestration |
4 |
0.4% |
context-engineering + eval-driven-development + human-in-the-loop-gate + tool-use-orchestration |
3 |
0.3% |
data-quality-loop + eval-driven-development + guardrails-io-filtering + human-in-the-loop-gate + llm-as-judge |
3 |
0.3% |
eval-driven-development + fine-tuning-adaptation + rag-retrieval |
3 |
0.3% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
135 |
13.2% |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024)
agentic-system-for-autonomous-code-monitoring-and-maintenance (ramp, 2026)
ai-agent-for-automated-quality-assurance-testing-in-cryptocurrency-platform (coinbase, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
102 |
10.0% |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026)
agent-driven-development-for-ai-research-using-github-copilot-cli (github, 2026)
agent-driven-ui-framework-migration-at-enterprise-scale (block, 2026) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
71 |
6.9% |
agentic-ai-architecture-for-investment-management-platform (blackrock, year n/a)
agentic-ai-for-legal-research-building-deep-research-in-westlaw-and-cocounsel (thomson_reuters, 2025)
agentic-code-reviewers-as-system-protectors (block, 2026) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
55 |
5.4% |
advanced-speaker-diarization-and-attributed-transcription-using-deep-learning-models (pyannoteai, 2026)
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025)
ai-driven-collateral-allocation-optimization-in-fintech (mercado_libre, 2025) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
42 |
4.1% |
ai-agent-for-automated-merchant-classification-correction (ramp, 2025)
ai-powered-clinical-documentation-with-multi-region-healthcare-compliance (heidi_health, 2025)
ai-powered-fraud-detection-using-mixture-of-experts-and-federated-learning (feedzai, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
39 |
3.8% |
ai-agents-for-data-labeling-and-infrastructure-maintenance-at-scale (plaid, 2025)
ai-orchestrated-code-review-system-at-scale (cloudflare, 2026)
ai-powered-fan-engagement-and-content-personalization-for-global-football-audiences (dfl_/_bundesliga, 2025) |
| evaluation-governed pipeline + guarded enterprise assistant + context/memory system |
35 |
3.4% |
ai-powered-consent-education-tool-for-preventing-gender-based-violence (override, 2026)
ai-powered-healthcare-building-reliable-care-agents-in-production (sword_health, 2025)
ai-ready-data-infrastructure-for-conversational-financial-analytics (vanguard, 2026) |
| evaluation-governed pipeline + domain-adapted model pipeline + cost/latency optimized serving |
29 |
2.8% |
ai-lab-a-pre-production-framework-for-ml-performance-testing-and-optimization (meta, 2024)
ai-powered-music-lyric-analysis-and-semantic-search-platform (lyriclens, 2024)
automated-inventory-counting-with-multimodal-llms-in-grocery-fulfillment (picnic, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 3649 (3.57 per raw project). Tradeoffs: 3286 (3.22 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Document standards in wikis and SOPs |
Written process docs depended on humans remembering and applying them, whereas skills can make the golden path easier to enforce inside the agent workflow. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Let Claude Code write generic PR descriptions |
The generic descriptions summarized code rather than intent, which was not useful enough for human or agent reviewers. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Optimize token spend early |
Intercom accepted high model spend during the adoption phase because the perceived throughput alpha was worth learning faster. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Plug AI devices into the legacy data house |
The platform needed re-architected wiring, interoperability, and modernized foundations before AI could scale safely. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Tradeoff |
speed and learning over early token-cost optimization |
The source frames high model spend as an investment while throughput was increasing. |
summary / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
skills and hooks embedded in Claude Code workflows over wiki/SOP guidance that humans must remember |
Intercom treated skills as upstream guardrails analogous to CI/CD, making standards easier to follow before code reached review. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
a create-PR skill plus LLM judging over accepting Claude Code's default code-summary descriptions |
Default descriptions lost the intent reviewers needed, so Intercom codified its PR-description expectations and measured quality over historical data. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
A tool-selection process that considered culture, licenses, cost restrictions, and worker skills over Picking technically capable tools from a feature checklist alone |
The implementation had to survive inside Novartis' operating model and compliance environment. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
Infrastructure, cost, and latency optimization
Deduped projects: 1010. Industries: 16. Provenance: 947 source-backed / 63 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
eval-driven-development |
473 |
46.8% |
tool-use-orchestration |
374 |
37.0% |
human-in-the-loop-gate |
307 |
30.4% |
observability-tracing |
301 |
29.8% |
rag-retrieval |
262 |
25.9% |
cost-aware-routing |
245 |
24.3% |
model-routing |
242 |
24.0% |
latency-optimization |
198 |
19.6% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
| none |
6 |
0.6% |
cost-aware-routing |
5 |
0.5% |
eval-driven-development + fine-tuning-adaptation + retrieval-ranking |
4 |
0.4% |
eval-driven-development + multi-agent-orchestration |
4 |
0.4% |
context-engineering + eval-driven-development |
3 |
0.3% |
cost-aware-routing + eval-driven-development + fine-tuning-adaptation |
3 |
0.3% |
cost-aware-routing + tool-use-orchestration |
3 |
0.3% |
eval-driven-development + tool-use-orchestration |
3 |
0.3% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
79 |
7.8% |
adopting-model-context-protocol-mcp-in-financial-services-for-ai-system-integration (evergreen_wealth_/bloomberg/_saxo_bank, 2026)
ai-agent-for-automated-quality-assurance-testing-in-cryptocurrency-platform (coinbase, 2025)
ai-agent-powered-compliance-review-automation-for-financial-services (stripe, 2024) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
58 |
5.7% |
accelerating-llm-inference-with-speculative-decoding-for-ai-agent-applications (linkedin, 2025)
advanced-speaker-diarization-and-attributed-transcription-using-deep-learning-models (pyannoteai, 2026)
agent-based-ai-assistants-for-enterprise-and-e-commerce-applications (prosus, 2024) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
53 |
5.2% |
ai-powered-clinical-documentation-with-multi-region-healthcare-compliance (heidi_health, 2025)
ai-powered-financial-assistant-for-automated-expense-management (brex, 2023)
ai-powered-food-image-generation-system-at-scale (delivery_hero, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
51 |
5.0% |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026)
agent-driven-ui-framework-migration-at-enterprise-scale (block, 2026)
agentic-ai-search-with-custom-evaluation-framework-for-church-management (pushpay, 2026) |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
44 |
4.4% |
ai-agents-for-data-labeling-and-infrastructure-maintenance-at-scale (plaid, 2025)
ai-orchestrated-code-review-system-at-scale (cloudflare, 2026)
ai-powered-call-center-agents-for-healthcare-operations (heyrevia, 2023) |
| retrieval-grounded assistant + evaluation-governed pipeline + cost/latency optimized serving |
42 |
4.2% |
ai-powered-crm-insights-with-rag-and-text-to-sql (tp_icap, 2025)
bert-based-sequence-models-for-contextual-product-recommendations (instacart, 2024)
building-a-comprehensive-llm-platform-for-food-delivery-services (swiggy, 2024) |
| evaluation-governed pipeline + domain-adapted model pipeline + cost/latency optimized serving |
36 |
3.6% |
ai-lab-a-pre-production-framework-for-ml-performance-testing-and-optimization (meta, 2024)
ai-powered-music-lyric-analysis-and-semantic-search-platform (lyriclens, 2024)
ai-powered-trust-and-safety-toolkit-with-custom-model-training-and-adaptive-moderation (musubi, 2026) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
33 |
3.3% |
agentic-ai-for-automated-absence-reporting-and-shift-management-at-airport-operations (manchester_airports_group, 2024)
agentic-platform-engineering-hub-for-cloud-operations-automation (thomson_reuters, 2026)
ai-agent-mode-and-model-context-protocol-for-autonomous-code-generation (github, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 3570 (3.53 per raw project). Tradeoffs: 3193 (3.16 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Document standards in wikis and SOPs |
Written process docs depended on humans remembering and applying them, whereas skills can make the golden path easier to enforce inside the agent workflow. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Let Claude Code write generic PR descriptions |
The generic descriptions summarized code rather than intent, which was not useful enough for human or agent reviewers. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Optimize token spend early |
Intercom accepted high model spend during the adoption phase because the perceived throughput alpha was worth learning faster. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Traditional lexical and embedding metrics |
They caught extreme drift but missed citation gaps, hallucinated filenames, formatting errors, and factual failures. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
speed and learning over early token-cost optimization |
The source frames high model spend as an investment while throughput was increasing. |
summary / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
skills and hooks embedded in Claude Code workflows over wiki/SOP guidance that humans must remember |
Intercom treated skills as upstream guardrails analogous to CI/CD, making standards easier to follow before code reached review. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
a create-PR skill plus LLM judging over accepting Claude Code's default code-summary descriptions |
Default descriptions lost the intent reviewers needed, so Intercom codified its PR-description expectations and measured quality over historical data. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
Automated evaluation gates on every material pipeline change over Faster ad-hoc experimentation without hard regression thresholds |
Dropbox accepted extra evaluation overhead because conversational AI regressions could appear in any stage of the pipeline. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
Context engineering and memory management
Deduped projects: 754. Industries: 16. Provenance: 717 source-backed / 37 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
tool-use-orchestration |
392 |
52.0% |
eval-driven-development |
378 |
50.1% |
human-in-the-loop-gate |
269 |
35.7% |
rag-retrieval |
250 |
33.2% |
observability-tracing |
209 |
27.7% |
multi-agent-orchestration |
175 |
23.2% |
safety-policy-enforcement |
156 |
20.7% |
structured-output-contract |
145 |
19.2% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
context-engineering + eval-driven-development |
4 |
0.5% |
eval-driven-development + fine-tuning-adaptation + retrieval-ranking |
4 |
0.5% |
eval-driven-development + tool-use-orchestration |
4 |
0.5% |
rag-retrieval + tool-use-orchestration |
4 |
0.5% |
eval-driven-development + human-in-the-loop-gate + rag-retrieval |
3 |
0.4% |
human-in-the-loop-gate + rag-retrieval |
3 |
0.4% |
multi-agent-orchestration + rag-retrieval |
3 |
0.4% |
safety-policy-enforcement + tool-use-orchestration |
3 |
0.4% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
90 |
11.9% |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026)
agent-driven-ui-framework-migration-at-enterprise-scale (block, 2026)
ai-agent-optimization-using-claude-to-systematically-improve-memory-extraction-quality (lerim, 2026) |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
82 |
10.9% |
ai-agent-solutions-for-data-warehouse-access-and-security (meta, 2025)
ai-agents-for-automated-product-quality-testing-and-bug-detection (coinbase, 2025)
ai-managed-services-and-agent-operations-at-enterprise-scale (pricewaterhousecooper, 2026) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
57 |
7.6% |
agentic-ai-for-automated-absence-reporting-and-shift-management-at-airport-operations (manchester_airports_group, 2024)
agentic-ai-for-legal-research-building-deep-research-in-westlaw-and-cocounsel (thomson_reuters, 2025)
agentic-code-reviewers-as-system-protectors (block, 2026) |
| retrieval-grounded assistant + multi-agent/tool workflow + context/memory system |
44 |
5.8% |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023)
agentic-ai-for-title-operations-workflow-optimization (rocket, 2026)
agentic-ai-system-for-document-summarization-and-analysis (moveworks, 2024) |
| retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline |
32 |
4.2% |
ai-powered-customer-service-and-call-center-transformation-with-multi-agent-systems (fastweb_/_vodafone, 2025)
ai-powered-multi-agent-decision-support-system-for-enterprise-strategic-planning (coinbase, 2025)
ai-powered-multi-agent-decision-support-system-for-strategic-business-decisions (coinbase, 2025) |
| multi-agent/tool workflow + context/memory system + cost/latency optimized serving |
30 |
4.0% |
agentic-ai-architecture-for-meeting-intelligence-and-productivity-automation (zoom, 2025)
ai-agents-for-documenting-tribal-knowledge-in-large-scale-data-pipelines (meta, 2026)
ai-employee-agent-operating-in-slack-with-multi-tool-integration (viktor, 2026) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
27 |
3.6% |
ai-powered-menu-description-generation-for-restaurant-platforms (doordash, 2025)
ai-powered-sales-assistant-for-go-to-market-team-productivity (openai, 2025)
automated-knowledge-base-enhancement-using-llms-and-clustering-for-customer-support (doordash, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
26 |
3.4% |
ai-powered-automated-issue-resolution-achieving-state-of-the-art-performance-on-swe-bench (trae, 2025)
ai-powered-medical-content-review-and-revision-at-scale (flo_health, 2026)
building-a-multi-model-llm-api-marketplace-and-infrastructure-platform (openrouter, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 2786 (3.69 per raw project). Tradeoffs: 2512 (3.33 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Document standards in wikis and SOPs |
Written process docs depended on humans remembering and applying them, whereas skills can make the golden path easier to enforce inside the agent workflow. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Let Claude Code write generic PR descriptions |
The generic descriptions summarized code rather than intent, which was not useful enough for human or agent reviewers. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Optimize token spend early |
Intercom accepted high model spend during the adoption phase because the perceived throughput alpha was worth learning faster. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Manual documentation by consultants |
Even after scaling to 12 consultants, the process was slow, expensive, inconsistent, and missed migration timelines. |
transcript / stated |
accelerating-sap-s-4hana-migration-and-custom-code-documentation-with-generative-ai (axfood_/_harman, 2025) |
| Tradeoff |
speed and learning over early token-cost optimization |
The source frames high model spend as an investment while throughput was increasing. |
summary / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
skills and hooks embedded in Claude Code workflows over wiki/SOP guidance that humans must remember |
Intercom treated skills as upstream guardrails analogous to CI/CD, making standards easier to follow before code reached review. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
a create-PR skill plus LLM judging over accepting Claude Code's default code-summary descriptions |
Default descriptions lost the intent reviewers needed, so Intercom codified its PR-description expectations and measured quality over historical data. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
Bedrock and Amazon Q Developer generated structured documentation over manual consultant documentation for every object |
The AI workflow produced uniform business and technical documentation much faster and at lower cost than the scaled manual plan. |
transcript / stated |
accelerating-sap-s-4hana-migration-and-custom-code-documentation-with-generative-ai (axfood_/_harman, 2025) |
Safety, compliance, and trust controls
Deduped projects: 688. Industries: 16. Provenance: 666 source-backed / 22 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
eval-driven-development |
321 |
46.7% |
human-in-the-loop-gate |
305 |
44.3% |
rag-retrieval |
268 |
39.0% |
tool-use-orchestration |
266 |
38.7% |
safety-policy-enforcement |
239 |
34.7% |
observability-tracing |
207 |
30.1% |
privacy-preserving-processing |
142 |
20.6% |
model-routing |
135 |
19.6% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
eval-driven-development + tool-use-orchestration |
5 |
0.7% |
data-quality-loop + eval-driven-development + human-in-the-loop-gate + observability-tracing + rag-retrieval |
4 |
0.6% |
rag-retrieval + safety-policy-enforcement + tool-use-orchestration |
4 |
0.6% |
safety-policy-enforcement + tool-use-orchestration |
4 |
0.6% |
eval-driven-development + human-in-the-loop-gate + llm-as-judge |
3 |
0.4% |
eval-driven-development + observability-tracing + safety-policy-enforcement |
3 |
0.4% |
human-in-the-loop-gate + rag-retrieval + safety-policy-enforcement |
3 |
0.4% |
multi-agent-orchestration + safety-policy-enforcement |
3 |
0.4% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
94 |
13.7% |
adopting-model-context-protocol-mcp-in-financial-services-for-ai-system-integration (evergreen_wealth_/bloomberg/_saxo_bank, 2026)
ai-agent-development-and-evaluation-platform-for-insurance-underwriting (snorkel, 2025)
ai-agent-for-automated-quality-assurance-testing-in-cryptocurrency-platform (coinbase, 2025) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
51 |
7.4% |
ai-agent-for-automated-merchant-classification-and-transaction-matching (ramp, 2025)
ai-powered-client-services-assistant-for-post-trade-services (london_stock_exchange_group, 2025)
ai-powered-merchant-classification-correction-agent (ramp, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + guarded enterprise assistant |
50 |
7.3% |
agentic-ai-for-cloud-migration-and-application-modernization-at-scale (commonwealth_bank_of_australia, 2025)
ai-assistant-integration-for-manufacturing-execution-system-mes (42q, 2025)
ai-driven-clinical-trial-transformation-with-next-generation-data-platform (novartis, 2025) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
47 |
6.8% |
ai-powered-betting-assistant-for-sports-wagering-platform (fanduel, 2025)
ai-powered-clinical-documentation-with-multi-region-healthcare-compliance (heidi_health, 2025)
ai-powered-financial-assistant-for-automated-expense-management (brex, 2023) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
40 |
5.8% |
agentic-ai-architecture-for-investment-management-platform (blackrock, year n/a)
agentic-ai-for-automated-absence-reporting-and-shift-management-at-airport-operations (manchester_airports_group, 2024)
agentic-platform-engineering-hub-for-cloud-operations-automation (thomson_reuters, 2026) |
| evaluation-governed pipeline + guarded enterprise assistant + context/memory system |
31 |
4.5% |
ai-powered-consent-education-tool-for-preventing-gender-based-violence (override, 2026)
ai-powered-contact-center-transformation-for-student-support-services (anthology, 2024)
ai-powered-healthcare-building-reliable-care-agents-in-production (sword_health, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
31 |
4.5% |
ai-agent-for-automated-root-cause-analysis-in-production-systems (cleric, 2025)
ai-agent-system-for-automated-security-investigation-and-alert-triage (slack, 2025)
ai-agents-for-life-sciences-rd-accelerating-drug-discovery-with-context-rich-data (benchling, 2026) |
| evaluation-governed pipeline + guarded enterprise assistant + domain-adapted model pipeline |
25 |
3.6% |
adversarial-grammatical-error-correction-at-scale-for-writing-assistance (grammarly, 2021)
ai-powered-marketing-compliance-automation-system (remitly, year n/a)
ai-powered-nutrition-guidance-with-fine-tuned-llama-models (omada_health, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 2442 (3.55 per raw project). Tradeoffs: 2181 (3.17 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Traditional lexical and embedding metrics |
They caught extreme drift but missed citation gaps, hallucinated filenames, formatting errors, and factual failures. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Rejected alternative |
Public benchmark datasets alone |
Public datasets lacked the long-tail phrasing and content mix of real Dash users. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Rejected alternative |
Spreadsheet-managed experiments |
Spreadsheets made results scattered, hard to reproduce, and hard to compare across model or prompt versions. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Rejected alternative |
Only offline or staging evaluation |
Dropbox needed asynchronous live-traffic scoring to catch silent degradation from real user queries without adding latency. |
article / implied |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
Automated evaluation gates on every material pipeline change over Faster ad-hoc experimentation without hard regression thresholds |
Dropbox accepted extra evaluation overhead because conversational AI regressions could appear in any stage of the pipeline. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
LLM judges with calibrated rubrics over BLEU, ROUGE, BERTScore, and embedding similarity alone |
Dropbox needed factuality, citation, formatting, and tone checks that traditional text metrics did not enforce. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
layered merge, stage, and production gates over one large evaluation pass late in release |
Layered gates let Dropbox block obvious regressions early and catch scale-only issues in production-like or live settings. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
A tool-selection process that considered culture, licenses, cost restrictions, and worker skills over Picking technically capable tools from a feature checklist alone |
The implementation had to survive inside Novartis' operating model and compliance environment. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
Customer support and conversational service
Deduped projects: 664. Industries: 15. Provenance: 626 source-backed / 38 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
tool-use-orchestration |
321 |
48.3% |
eval-driven-development |
300 |
45.2% |
human-in-the-loop-gate |
242 |
36.4% |
rag-retrieval |
242 |
36.4% |
observability-tracing |
217 |
32.7% |
safety-policy-enforcement |
169 |
25.5% |
multi-agent-orchestration |
143 |
21.5% |
model-routing |
131 |
19.7% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
safety-policy-enforcement + tool-use-orchestration |
4 |
0.6% |
data-quality-loop + eval-driven-development + human-in-the-loop-gate + observability-tracing + rag-retrieval |
3 |
0.5% |
eval-driven-development + human-in-the-loop-gate + observability-tracing + rag-retrieval + safety-policy-enforcement |
3 |
0.5% |
eval-driven-development + human-in-the-loop-gate + rag-retrieval |
3 |
0.5% |
eval-driven-development + llm-as-judge |
3 |
0.5% |
eval-driven-development + multi-agent-orchestration |
3 |
0.5% |
human-in-the-loop-gate + rag-retrieval |
3 |
0.5% |
multi-agent-orchestration + tool-use-orchestration |
3 |
0.5% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
79 |
11.9% |
adopting-model-context-protocol-mcp-in-financial-services-for-ai-system-integration (evergreen_wealth_/bloomberg/saxo_bank, 2026)
ai-agents-and-intelligent-observability-for-devops-modernization (hrs_group/netflix/_harness, 2026)
ai-assisted-activity-onboarding-in-travel-marketplace (getyourguide, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + guarded enterprise assistant |
42 |
6.3% |
agentic-ai-systems-for-drug-discovery-and-business-intelligence (loka, 2025)
agentic-data-analyst-for-enterprise-analytics (ramp, 2025)
ai-assistant-integration-for-manufacturing-execution-system-mes (42q, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
40 |
6.0% |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026)
agent-testing-and-evaluation-using-autonomous-vehicle-simulation-principles (coval, 2023)
ai-agent-for-automated-root-cause-analysis-in-production-systems (cleric, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
39 |
5.9% |
agentic-ai-architecture-for-investment-management-platform (blackrock, year n/a)
agentic-ai-for-automated-absence-reporting-and-shift-management-at-airport-operations (manchester_airports_group, 2024)
ai-powered-chatbot-automation-with-hybrid-nlu-and-llm-approach (scotiabank, 2022) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
38 |
5.7% |
ai-agent-for-automated-merchant-classification-and-transaction-matching (ramp, 2025)
ai-powered-client-services-assistant-for-post-trade-services (london_stock_exchange_group, 2025)
ai-powered-insurance-claims-chatbot-with-continuous-feedback-loop (allianz, 2023) |
| retrieval-grounded assistant + multi-agent/tool workflow + context/memory system |
30 |
4.5% |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023)
agent-first-ai-development-platform-with-multi-surface-orchestration (google_deepmind, 2025)
agentic-ai-for-title-operations-workflow-optimization (rocket, 2026) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
26 |
3.9% |
accelerating-llm-inference-with-speculative-decoding-for-ai-agent-applications (linkedin, 2025)
agent-based-ai-assistants-for-enterprise-and-e-commerce-applications (prosus, 2024)
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
26 |
3.9% |
ai-powered-call-center-agents-for-healthcare-operations (heyrevia, 2023)
ai-powered-on-call-assistant-for-airflow-pipeline-debugging (wix, 2026)
build-vs-buy-ai-agents-enterprise-deployment-lessons-from-1-000-companies (dust, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 2375 (3.58 per raw project). Tradeoffs: 2131 (3.21 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Document standards in wikis and SOPs |
Written process docs depended on humans remembering and applying them, whereas skills can make the golden path easier to enforce inside the agent workflow. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Let Claude Code write generic PR descriptions |
The generic descriptions summarized code rather than intent, which was not useful enough for human or agent reviewers. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Optimize token spend early |
Intercom accepted high model spend during the adoption phase because the perceived throughput alpha was worth learning faster. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Rejected alternative |
Traditional lexical and embedding metrics |
They caught extreme drift but missed citation gaps, hallucinated filenames, formatting errors, and factual failures. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
| Tradeoff |
speed and learning over early token-cost optimization |
The source frames high model spend as an investment while throughput was increasing. |
summary / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
skills and hooks embedded in Claude Code workflows over wiki/SOP guidance that humans must remember |
Intercom treated skills as upstream guardrails analogous to CI/CD, making standards easier to follow before code reached review. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
a create-PR skill plus LLM judging over accepting Claude Code's default code-summary descriptions |
Default descriptions lost the intent reviewers needed, so Intercom codified its PR-description expectations and measured quality over historical data. |
transcript / stated |
2x-engineering-throughput-through-ai-first-development-platform (intercom, 2026) |
| Tradeoff |
Automated evaluation gates on every material pipeline change over Faster ad-hoc experimentation without hard regression thresholds |
Dropbox accepted extra evaluation overhead because conversational AI regressions could appear in any stage of the pipeline. |
article / stated |
a-practical-blueprint-for-evaluating-conversational-ai-at-scale (dropbox, 2025) |
Document processing and structured extraction
Deduped projects: 391. Industries: 16. Provenance: 363 source-backed / 28 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
eval-driven-development |
170 |
43.5% |
human-in-the-loop-gate |
160 |
40.9% |
tool-use-orchestration |
138 |
35.3% |
structured-output-contract |
129 |
33.0% |
rag-retrieval |
125 |
32.0% |
observability-tracing |
86 |
22.0% |
model-routing |
79 |
20.2% |
cost-aware-routing |
69 |
17.6% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
| none |
5 |
1.3% |
context-engineering + eval-driven-development |
4 |
1.0% |
context-engineering + eval-driven-development + human-in-the-loop-gate + tool-use-orchestration |
3 |
0.8% |
human-in-the-loop-gate + multi-agent-orchestration |
3 |
0.8% |
batch-offline-processing + eval-driven-development + model-routing + structured-output-contract |
2 |
0.5% |
canary-rollout + latency-optimization + observability-tracing + tool-use-orchestration |
2 |
0.5% |
cost-aware-routing + fallback-degradation + latency-optimization + model-routing + observability-tracing |
2 |
0.5% |
cost-aware-routing + guardrails-io-filtering + prompt-chaining + structured-output-contract |
2 |
0.5% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
32 |
8.2% |
ai-agent-development-and-evaluation-platform-for-insurance-underwriting (snorkel, 2025)
ai-assisted-activity-onboarding-in-travel-marketplace (getyourguide, 2025)
ai-driven-multi-agent-system-for-dynamic-product-taxonomy-evolution (shopify, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
27 |
6.9% |
agentic-ai-architecture-for-investment-management-platform (blackrock, year n/a)
agentic-ai-for-legal-research-building-deep-research-in-westlaw-and-cocounsel (thomson_reuters, 2025)
agentic-ai-system-for-construction-industry-tender-management-and-quote-generation (tendos_ai, 2026) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
19 |
4.9% |
ai-powered-menu-description-generation-for-restaurant-platforms (doordash, 2025)
ai-powered-vehicle-information-platform-for-dealership-sales-support (toyota, 2025)
building-a-generic-recommender-system-api-with-privacy-first-design (slack, 2023) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
18 |
4.6% |
automated-software-development-insights-and-communication-platform (blueprint_ai, 2023)
autonomous-bug-investigation-and-resolution-agent (basis, 2026)
autonomous-multi-agent-system-for-complex-software-development (factory, 2026) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
18 |
4.6% |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025)
ai-driven-collateral-allocation-optimization-in-fintech (mercado_libre, 2025)
ai-powered-automated-issue-resolution-achieving-state-of-the-art-performance-on-swe-bench (trae, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
17 |
4.3% |
ai-powered-identity-verification-and-fraud-detection-for-online-lending (sun_finance, 2026)
ai-powered-natural-language-search-for-vehicle-marketplace (coches.net, 2024)
ai-powered-tour-guide-for-financial-platform-navigation (ramp, 2024) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
16 |
4.1% |
ai-powered-financial-assistant-for-automated-expense-management (brex, 2023)
building-ai-assist-llm-integration-for-e-commerce-product-listings (mercari, 2023)
evaluating-long-context-performance-in-legal-ai-applications (thomson_reuters, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline |
15 |
3.8% |
ai-powered-content-generation-and-shot-commentary-system-for-live-golf-tournament-coverage (pga_tour, 2025)
ai-powered-shift-left-testing-platform-with-multiple-llm-agents (qyrusai, 2025)
automated-esg-reporting-with-agentic-ai-for-enterprise-sustainability-compliance (gardenia_technologies, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 1355 (3.47 per raw project). Tradeoffs: 1213 (3.10 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Ask the model for the final artifact in one pass |
Room for Thought and Monte Carlo split intermediate reasoning or options before producing the final answer. |
article / implied |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023) |
| Rejected alternative |
Let the model freely choose classifier labels |
The internal proxy endpoint guarantees valid output and can use logit bias to nudge valid labels. |
article / stated |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023) |
| Rejected alternative |
Require the model to start in exactly the desired format unaided |
Puppetry reduces formatting burden by telling the assistant it has already started responding. |
article / stated |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023) |
| Rejected alternative |
Chatbot-only agent interface |
Otto wanted agents to run in the tabular interface business users already understand. |
transcript / stated |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
| Tradeoff |
intermediate outlines, multiple samples, and self-correction over single-pass final generation |
Instacart found that forcing intermediate structure or sampling several options improved the final output for complex writing tasks. |
article / stated |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023) |
| Tradeoff |
a proxy endpoint with constrained valid output and logit bias over unconstrained natural-language classification |
Production code benefits from guaranteed label shape, while logit bias nudges ambiguous choices toward allowed outputs. |
article / stated |
advanced-prompt-engineering-techniques-for-production-llm-applications (instacart, 2023) |
| Tradeoff |
spreadsheet UX over chatbot-only interface |
The summary says the familiar spreadsheet UX lets finance/accounting users harness agents without technical expertise. |
summary / stated |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
| Tradeoff |
cell-level agents inside spreadsheets over a chatbot or database-style agent interface |
The spreadsheet lets agents fit the business-user workflow instead of forcing users into technical abstractions. |
transcript / stated |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
Agentic workflow automation and orchestration
Deduped projects: 338. Industries: 15. Provenance: 319 source-backed / 19 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
tool-use-orchestration |
232 |
68.6% |
multi-agent-orchestration |
162 |
47.9% |
human-in-the-loop-gate |
149 |
44.1% |
eval-driven-development |
138 |
40.8% |
observability-tracing |
121 |
35.8% |
agent-memory |
79 |
23.4% |
safety-policy-enforcement |
79 |
23.4% |
model-routing |
57 |
16.9% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
eval-driven-development + tool-use-orchestration |
3 |
0.9% |
human-in-the-loop-gate + multi-agent-orchestration |
3 |
0.9% |
agent-memory + model-routing + tool-use-orchestration |
2 |
0.6% |
agent-memory + multi-agent-orchestration + observability-tracing + safety-policy-enforcement + tool-use-orchestration |
2 |
0.6% |
batch-offline-processing + human-in-the-loop-gate + latency-optimization + multi-agent-orchestration + tool-use-orchestration |
2 |
0.6% |
context-engineering + multi-agent-orchestration + tool-use-orchestration |
2 |
0.6% |
data-quality-loop + human-in-the-loop-gate |
2 |
0.6% |
durable-execution + eval-driven-development + human-in-the-loop-gate + multi-agent-orchestration + observability-tracing |
2 |
0.6% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
67 |
19.8% |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024)
agentic-system-for-autonomous-code-monitoring-and-maintenance (ramp, 2026)
ai-agent-development-and-evaluation-platform-for-insurance-underwriting (snorkel, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
46 |
13.6% |
ai-agent-system-for-automated-security-investigation-and-alert-triage (slack, 2025)
ai-agents-for-accelerating-model-development-and-framework-migration (linkedin, 2026)
ai-driven-incident-response-and-automated-remediation-for-digital-media-platform (iheart, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
43 |
12.7% |
agentic-ai-for-automated-absence-reporting-and-shift-management-at-airport-operations (manchester_airports_group, 2024)
agentic-ai-for-legal-research-building-deep-research-in-westlaw-and-cocounsel (thomson_reuters, 2025)
agentic-ai-system-for-construction-industry-tender-management-and-quote-generation (tendos_ai, 2026) |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
22 |
6.5% |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025)
ai-agents-for-ml-experiment-orchestration-reducing-friction-in-machine-learning-workflows (teads, 2026)
ai-driven-collateral-allocation-optimization-in-fintech (mercado_libre, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
17 |
5.0% |
ai-agents-for-data-labeling-and-infrastructure-maintenance-at-scale (plaid, 2025)
ai-powered-fan-engagement-and-content-personalization-for-global-football-audiences (dfl_/_bundesliga, 2025)
ai-powered-on-call-assistant-for-airflow-pipeline-debugging (wix, 2026) |
| multi-agent/tool workflow + context/memory system + cost/latency optimized serving |
15 |
4.4% |
ai-agents-for-documenting-tribal-knowledge-in-large-scale-data-pipelines (meta, 2026)
ai-employee-agent-operating-in-slack-with-multi-tool-integration (viktor, 2026)
ai-powered-code-review-system-at-scale (uber, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + context/memory system |
14 |
4.1% |
agent-first-ai-development-platform-with-multi-surface-orchestration (google_deepmind, 2025)
agentic-ai-for-title-operations-workflow-optimization (rocket, 2026)
agentic-ai-systems-for-legal-tax-and-compliance-workflows (thomson_reuters, 2024) |
| multi-agent/tool workflow + durable async execution + context/memory system |
12 |
3.6% |
agentic-mapreduce-for-whole-codebase-security-scanning (devin, 2026)
ai-native-workflow-transformation-for-enterprise-renewals-at-scale (cisco, 2026)
ai-powered-chief-of-staff-scaling-agent-architecture-from-monolith-to-distributed-system (outropy, 2024) |
Decisions and Rejected Alternatives
Extracted alternatives: 1219 (3.61 per raw project). Tradeoffs: 1098 (3.25 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Monitor individual LLM API requests |
Single-call monitoring missed the multi-step and tool-using flow of an agentic software-development system. |
article / stated |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024) |
| Rejected alternative |
Scroll manually through full traces |
Hundreds-step traces required search and filtering inside the trace to debug specific events. |
article / stated |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024) |
| Rejected alternative |
Chatbot-only agent interface |
Otto wanted agents to run in the tabular interface business users already understand. |
transcript / stated |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
| Rejected alternative |
Always use the strongest model |
Latency and cost make heavy reasoning models a poor fit for many routine tasks. |
transcript / stated |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
| Tradeoff |
End-to-end execution tracing across planning, code, dependencies, and deployment steps over Per-request LLM monitoring that only sees isolated provider calls |
Replit Agent's hundreds-step workflows required holistic visibility to debug where users got stuck or intervention was needed. |
article / stated |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024) |
| Tradeoff |
full agent execution traces over individual LLM API-call monitoring |
Replit Agent problems emerge across planning, tools, code execution, and parallel threads, so request-level views lack enough context. |
article / stated |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024) |
| Tradeoff |
large-trace performance plus intra-trace search and thread views over linear inspection of every trace step |
The team needed to find specific events and correlate related parallel traces without manually reading hundreds of steps. |
article / stated |
advanced-agent-monitoring-and-debugging-with-langsmith-integration (replit, 2024) |
| Tradeoff |
spreadsheet UX over chatbot-only interface |
The summary says the familiar spreadsheet UX lets finance/accounting users harness agents without technical expertise. |
summary / stated |
agent-based-workflow-automation-in-spreadsheets-for-non-technical-users (otto, 2025) |
Domain-specific model adaptation
Deduped projects: 67. Industries: 14. Provenance: 61 source-backed / 6 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
fine-tuning-adaptation |
57 |
85.1% |
eval-driven-development |
32 |
47.8% |
cost-aware-routing |
22 |
32.8% |
human-in-the-loop-gate |
18 |
26.9% |
data-quality-loop |
17 |
25.4% |
latency-optimization |
17 |
25.4% |
observability-tracing |
13 |
19.4% |
tool-use-orchestration |
10 |
14.9% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
cost-aware-routing + data-quality-loop + fine-tuning-adaptation + privacy-preserving-processing |
2 |
3.0% |
data-quality-loop + eval-driven-development + fine-tuning-adaptation + human-in-the-loop-gate + observability-tracing |
2 |
3.0% |
eval-driven-development + fine-tuning-adaptation + retrieval-ranking |
2 |
3.0% |
batch-offline-processing + canary-rollout + durable-execution + eval-driven-development + fine-tuning-adaptation |
1 |
1.5% |
batch-offline-processing + cost-aware-routing + data-quality-loop + fine-tuning-adaptation + latency-optimization |
1 |
1.5% |
batch-offline-processing + cost-aware-routing + durable-execution + observability-tracing |
1 |
1.5% |
batch-offline-processing + cost-aware-routing + eval-driven-development + fine-tuning-adaptation + retrieval-ranking |
1 |
1.5% |
batch-offline-processing + cost-aware-routing + fine-tuning-adaptation + latency-optimization + retrieval-ranking |
1 |
1.5% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| evaluation-governed pipeline + domain-adapted model pipeline + cost/latency optimized serving |
11 |
16.4% |
automating-radiology-report-generation-with-fine-tuned-llms (heidelberg_university, 2024)
domain-specific-model-training-and-reinforcement-learning-with-verifiable-rewards-for-qa-tasks (ramp, 2026)
fine-tuning-and-deploying-llms-for-customer-service-contact-centers (swisscom, 2025) |
| guarded enterprise assistant + domain-adapted model pipeline + cost/latency optimized serving |
8 |
11.9% |
auto-generated-document-summaries-using-abstractive-summarization (google, 2022)
building-a-custom-vision-llm-for-document-processing-at-scale (grab, 2025)
developing-and-deploying-domain-adapted-llms-for-e-commerce-through-continued-pre-training (ebay, 2025) |
| evaluation-governed pipeline + guarded enterprise assistant + domain-adapted model pipeline |
7 |
10.4% |
ai-powered-contact-center-copilot-from-research-to-enterprise-scale-production (cresta_/_openai, 2025)
building-production-ready-healthcare-ai-that-scales-with-model-progress (anterior, 2026)
llm-integration-for-customer-support-automation-and-enhancement (airbnb, 2022) |
| multi-agent/tool workflow + evaluation-governed pipeline + domain-adapted model pipeline |
5 |
7.5% |
building-foundation-models-for-computer-use-agents (tzafon, 2026)
fine-tuning-llms-for-multi-agent-orchestration-in-code-generation (cosine, 2025)
formal-verification-and-verified-ai-for-mathematical-reasoning-at-scale (axiom_math, 2025) |
| retrieval-grounded assistant + evaluation-governed pipeline + domain-adapted model pipeline |
4 |
6.0% |
building-price-prediction-and-similar-item-search-models-for-e-commerce (ebay, 2024)
foundation-model-for-large-scale-personalized-recommendation (netflix, 2025)
foundation-model-for-unified-personalization-at-scale (netflix, 2024) |
| domain-adapted model pipeline + context/memory system + cost/latency optimized serving |
3 |
4.5% |
optimizing-gpu-memory-usage-in-llm-training-with-liger-kernel (linkedin, 2025)
semi-supervised-fine-tuning-of-compact-vision-language-models-for-product-attribute-extraction (flipkart, 2025)
training-low-resource-language-models-with-custom-tokenization-and-kernel-optimization (azercell, 2026) |
| domain-adapted model pipeline + durable async execution + cost/latency optimized serving |
3 |
4.5% |
building-and-deploying-a-code-generation-llm-at-scale (replit, 2024)
integrating-foundation-models-into-production-personalization-systems (netflix, 2025)
small-language-models-for-real-time-ad-relevance-prediction (doordash, 2026) |
| evaluation-governed pipeline + domain-adapted model pipeline + context/memory system |
3 |
4.5% |
domain-adaptation-of-llms-for-enterprise-use-through-multi-task-fine-tuning (wix, 2024)
ml-based-comment-ranker-for-llm-code-review-quality-improvement (atlassian, year n/a)
variable-aggression-code-autocomplete-with-fine-tuned-llms (windsurf, 2026) |
Decisions and Rejected Alternatives
Extracted alternatives: 230 (3.43 per raw project). Tradeoffs: 207 (3.09 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Prompt engineering and RAG only |
The post says high-stakes applications often still need advanced fine-tuning to reach production performance. |
article / implied |
advanced-fine-tuning-techniques-for-multi-agent-orchestration-at-scale (amazon, 2026) |
| Rejected alternative |
Use PPO for preference alignment by default |
DPO simplified alignment by removing the explicit reward model and working directly with preference pairs. |
article / stated |
advanced-fine-tuning-techniques-for-multi-agent-orchestration-at-scale (amazon, 2026) |
| Rejected alternative |
Apply the most advanced reasoning optimization everywhere |
Amazon A+ showed a classification task could use feature-based fine-tuning rather than GRPO or DAPO. |
article / stated |
advanced-fine-tuning-techniques-for-multi-agent-orchestration-at-scale (amazon, 2026) |
| Rejected alternative |
Dictionary matching with sentiment lexicons |
Limited vocabulary cannot handle contextual market meanings, such as a phrase that looks negative but is bullish. |
transcript / stated |
agentic-news-analysis-platform-for-digital-asset-market-making (fsi, 2025) |
| Tradeoff |
advanced fine-tuning over prompting/RAG-only implementation |
The source says about one in four enterprise use cases still need advanced fine-tuning. |
article / stated |
advanced-fine-tuning-techniques-for-multi-agent-orchestration-at-scale (amazon, 2026) |
| Tradeoff |
advanced fine-tuning for high-stakes use cases over prompting, RAG, and turnkey agents alone |
Amazon presents fine-tuning as necessary when stakes, domain specificity, governance, or complex process integration demand production-grade performance. |
article / stated |
advanced-fine-tuning-techniques-for-multi-agent-orchestration-at-scale (amazon, 2026) |
| Tradeoff |
match technique complexity to task requirements over using the newest reinforcement method for every agent component |
Classification tasks can use simpler feature-based tuning, while complex planning and reasoning may justify GRPO, DAPO, or related methods. |
article / stated |
advanced-fine-tuning-techniques-for-multi-agent-orchestration-at-scale (amazon, 2026) |
| Tradeoff |
embedding-based deduplication with fine-tuned BGE-m3 and OpenSearch collections over sending every duplicate report through the LLM |
Repeated reports would burn cost and compound latency in a trading workflow where milliseconds matter. |
transcript / stated |
agentic-news-analysis-platform-for-digital-asset-market-making (fsi, 2025) |
Multimodal, edge, and voice AI systems
Deduped projects: 67. Industries: 12. Provenance: 62 source-backed / 5 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
fine-tuning-adaptation |
21 |
31.3% |
model-routing |
21 |
31.3% |
tool-use-orchestration |
21 |
31.3% |
human-in-the-loop-gate |
20 |
29.9% |
eval-driven-development |
19 |
28.4% |
latency-optimization |
19 |
28.4% |
data-quality-loop |
15 |
22.4% |
cost-aware-routing |
14 |
20.9% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
eval-driven-development + fine-tuning-adaptation + latency-optimization + observability-tracing + structured-output-contract |
2 |
3.0% |
agent-memory + cost-aware-routing + latency-optimization + rag-retrieval |
1 |
1.5% |
agent-memory + cost-aware-routing + safety-policy-enforcement |
1 |
1.5% |
async-message-orchestration + eval-driven-development + fallback-degradation + human-in-the-loop-gate + latency-optimization |
1 |
1.5% |
async-message-orchestration + guardrails-io-filtering + human-in-the-loop-gate + model-routing |
1 |
1.5% |
batch-offline-processing + context-engineering + data-quality-loop + eval-driven-development + structured-output-contract |
1 |
1.5% |
batch-offline-processing + cost-aware-routing + durable-execution + observability-tracing |
1 |
1.5% |
batch-offline-processing + cost-aware-routing + latency-optimization |
1 |
1.5% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
7 |
10.4% |
ai-powered-call-center-agents-for-healthcare-operations (heyrevia, 2023)
ai-powered-fan-engagement-and-content-personalization-for-global-football-audiences (dfl_/_bundesliga, 2025)
ai-powered-video-analysis-and-highlight-generation-platform (accenture, 2025) |
| evaluation-governed pipeline + domain-adapted model pipeline + cost/latency optimized serving |
4 |
6.0% |
automated-inventory-counting-with-multimodal-llms-in-grocery-fulfillment (picnic, 2025)
automating-radiology-report-generation-with-fine-tuned-llms (heidelberg_university, 2024)
building-a-voice-assistant-from-open-source-llms-a-home-project-case-study (weights_&_biases, 2023) |
| guarded enterprise assistant + domain-adapted model pipeline + cost/latency optimized serving |
4 |
6.0% |
ai-powered-content-moderation-at-platform-scale (roblox, 2025)
ai-powered-content-moderation-at-scale-safechat-platform (doordash, 2025)
multi-company-panel-on-production-llm-deployment-strategies-and-small-language-model-optimization (meta_/aws/nvidia/_conversenow, 2024) |
| domain-adapted model pipeline + context/memory system + cost/latency optimized serving |
3 |
4.5% |
building-a-global-product-catalogue-with-multimodal-llms-at-scale (shopify, 2025)
optimizing-production-vision-pipelines-for-planet-image-generation (prem_ai, 2024)
semi-supervised-fine-tuning-of-compact-vision-language-models-for-product-attribute-extraction (flipkart, 2025) |
| guarded enterprise assistant + durable async execution + cost/latency optimized serving |
3 |
4.5% |
ai-generated-trip-reports-for-outdoor-recreation-guides (guidesly, 2026)
ai-powered-search-and-agent-automation-for-digital-asset-management (bynder, 2026)
building-a-production-ready-ai-phone-call-assistant-with-multi-modal-processing (realchar, 2023) |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
3 |
4.5% |
automated-image-generation-for-e-commerce-categories-using-multimodal-llms (ebay, 2025)
building-a-reliable-ai-quote-generation-assistant-with-langgraph (tradestack, 2024)
scaling-order-processing-automation-using-modular-llm-architecture (choco, 2025) |
| domain-adapted model pipeline + cost/latency optimized serving |
2 |
3.0% |
ai-powered-audio-enhancement-for-tv-and-movie-dialogue-clarity (amazon, 2025)
automated-sign-language-translation-using-large-language-models (vsl_labs, 2024) |
| evaluation-governed pipeline + durable async execution + cost/latency optimized serving |
2 |
3.0% |
building-a-production-ai-translation-and-lip-sync-system-at-scale (meta, 2023)
climate-tech-foundation-models-for-environmental-ai-applications (various, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 218 (3.25 per raw project). Tradeoffs: 180 (2.69 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Assign speakers to words with a simple timestamp join |
STT and diarization timestamps disagree, STT misses overlaps, and diarization may detect speech the transcriber does not. |
transcript / implied |
advanced-speaker-diarization-and-attributed-transcription-using-deep-learning-models (pyannoteai, 2026) |
| Rejected alternative |
Use generic single-speaker STT as-is |
Single-speaker training does not generalize cleanly to overlap, crosstalk, distant microphones, and speaker changes. |
transcript / stated |
advanced-speaker-diarization-and-attributed-transcription-using-deep-learning-models (pyannoteai, 2026) |
| Rejected alternative |
Keep agents inside the IDE/chat surface |
The platform treats agent work as spanning editor, browser, and manager surfaces. |
transcript / stated |
agent-first-ai-development-platform-with-multi-surface-orchestration (google_deepmind, 2025) |
| Rejected alternative |
Supervise agents through conversation logs |
Raw token streams are hard to follow for many concurrent or long-running tasks. |
transcript / implied |
agent-first-ai-development-platform-with-multi-surface-orchestration (google_deepmind, 2025) |
| Tradeoff |
a dedicated reconciliation layer over direct timestamp alignment |
Overlaps, missing transcripts, and shifted timestamps make direct assignment unreliable. |
transcript / stated |
advanced-speaker-diarization-and-attributed-transcription-using-deep-learning-models (pyannoteai, 2026) |
| Tradeoff |
Fast handoff from agent manager back into the editor with files, artifacts, and conversation preserved over Forcing every task to remain fully delegated to the agent |
The product recognized that some tasks still required direct developer control because trust in models was not universal. |
transcript / stated |
agent-first-ai-development-platform-with-multi-surface-orchestration (google_deepmind, 2025) |
| Tradeoff |
artifact-centered agent manager over raw conversation and token streams |
Artifacts present plans, progress, diagrams, and proof of work in forms that humans can review more easily than raw logs. |
transcript / stated |
agent-first-ai-development-platform-with-multi-surface-orchestration (google_deepmind, 2025) |
| Tradeoff |
a disposable AI strategy with replaceable services over model-specific end-to-end workflow rewrites |
Treating models as disposable increases up-front platform discipline but makes model replacement and prompt/version changes cheaper. |
transcript / stated |
ai-driven-media-analysis-and-content-assembly-platform-for-large-scale-video-archives (bloomberg_media, 2024) |
Data analytics and decision intelligence
Deduped projects: 57. Industries: 10. Provenance: 51 source-backed / 6 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
tool-use-orchestration |
34 |
59.6% |
human-in-the-loop-gate |
20 |
35.1% |
rag-retrieval |
19 |
33.3% |
eval-driven-development |
15 |
26.3% |
structured-output-contract |
15 |
26.3% |
multi-agent-orchestration |
12 |
21.1% |
observability-tracing |
11 |
19.3% |
safety-policy-enforcement |
11 |
19.3% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
latency-optimization + safety-policy-enforcement + structured-output-contract + tool-use-orchestration |
2 |
3.5% |
agent-memory + confidence-thresholding + eval-driven-development + guardrails-io-filtering + human-in-the-loop-gate |
1 |
1.8% |
agent-memory + context-compaction + guardrails-io-filtering + human-in-the-loop-gate + multi-agent-orchestration |
1 |
1.8% |
agent-memory + guardrails-io-filtering + latency-optimization + multi-agent-orchestration + rag-retrieval |
1 |
1.8% |
agent-memory + human-in-the-loop-gate + observability-tracing + tool-use-orchestration |
1 |
1.8% |
agent-memory + observability-tracing + rag-retrieval + tool-use-orchestration |
1 |
1.8% |
async-message-orchestration + cost-aware-routing + observability-tracing + privacy-preserving-processing |
1 |
1.8% |
batch-offline-processing + context-engineering + model-routing + privacy-preserving-processing + prompt-versioning |
1 |
1.8% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| retrieval-grounded assistant + multi-agent/tool workflow + guarded enterprise assistant |
7 |
12.3% |
agentic-data-analyst-for-enterprise-analytics (ramp, 2025)
building-and-deploying-production-ai-agents-for-enterprise-data-analysis (asterrave, 2024)
building-production-ready-ai-agents-for-internal-workflow-automation (vercel, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + context/memory system |
6 |
10.5% |
building-production-ready-sql-and-charting-agents-with-rag-integration (numbers_station, year n/a)
evolution-of-hermes-v3-building-a-conversational-ai-data-analyst (swiggy, 2024)
generative-ai-assistant-for-agricultural-field-trial-analysis (agmatix, 2024) |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
5 |
8.8% |
ai-powered-postmortem-analysis-for-site-reliability-engineering (zalando, 2025)
ai-powered-self-remediation-loop-for-large-scale-kubernetes-operations (salesforce, 2025)
automating-root-cause-analysis-using-amazon-bedrock-agents (bmw, 2025) |
| multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving |
5 |
8.8% |
building-a-natural-language-business-intelligence-interface-with-mcp (ramp, 2025)
building-a-natural-language-interface-to-business-spend-data-via-mcp-and-sql (ramp, 2025)
llm-powered-upskilling-assistant-in-steel-manufacturing (gerdau, 2024) |
| multi-agent/tool workflow + guarded enterprise assistant + context/memory system |
4 |
7.0% |
ai-agent-for-self-service-business-intelligence-with-text-to-sql (bgl, 2026)
building-production-ready-ai-analytics-agents-through-advanced-prompt-engineering (explai, 2025)
prosus-x61bbbd495b (prosus, 2024) |
| retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline |
3 |
5.3% |
ai-assisted-database-debugging-platform-at-scale (databricks, 2024)
building-a-collaborative-multi-agent-ai-ecosystem-for-enterprise-knowledge-access (doordash, 2025)
text-to-sql-solution-for-data-democratization-in-food-delivery-operations (swiggy, 2024) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
2 |
3.5% |
ai-powered-market-surveillance-system-for-financial-compliance (london_stock_exchange_group, 2025)
building-production-grade-agentic-ai-analytics-lessons-from-real-world-deployment (tellius, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + context/memory system |
2 |
3.5% |
building-production-analytics-agents-with-semantic-layer-integration (wobby, 2025)
building-robust-evaluation-systems-for-github-copilot (github, 2023) |
Decisions and Rejected Alternatives
Extracted alternatives: 199 (3.49 per raw project). Tradeoffs: 179 (3.14 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Plug AI devices into the legacy data house |
The platform needed re-architected wiring, interoperability, and modernized foundations before AI could scale safely. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Rejected alternative |
Choose tools only by feature fit |
Novartis had to account for culture, licenses, cost restrictions, and worker skills in a regulated operating model. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Rejected alternative |
Keep logs and audit traces in each source system |
Central observability avoided fragmented dashboard connections and gave users accountability over quality, access, security, speed, and cost. |
transcript / implied |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Rejected alternative |
Lock into proprietary or narrow platform capabilities |
Open standards were needed for GXP interoperability across sources, therapeutic areas, and multiple platforms. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Tradeoff |
A tool-selection process that considered culture, licenses, cost restrictions, and worker skills over Picking technically capable tools from a feature checklist alone |
The implementation had to survive inside Novartis' operating model and compliance environment. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Tradeoff |
a central observability platform over leaving logs, audit traces, and financial information distributed at source |
The team preferred central dashboards and accountability over many dashboard connections back to individual systems. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Tradeoff |
open standards and interoperable tools over platform-specific capabilities that required workarounds |
The GXP environment needed tools and AWS services that could communicate across sources, therapeutic areas, and platforms. |
transcript / stated |
accelerating-drug-development-with-ai-powered-clinical-trial-transformation (novartis, 2025) |
| Tradeoff |
Smart rounds with real-time sentiment analysis over Delayed HCAHPS-only analysis after discharge |
Earlier detection allowed staff to prioritize patients at risk of bad experiences before the feedback window closed. |
transcript / stated |
advancing-patient-experience-and-business-operations-analytics-with-generative-ai-in-healthcare (huron, 2025) |
Search, ranking, and recommendation quality
Deduped projects: 50. Industries: 6. Provenance: 47 source-backed / 3 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
eval-driven-development |
27 |
54.0% |
rag-retrieval |
16 |
32.0% |
cost-aware-routing |
14 |
28.0% |
human-in-the-loop-gate |
13 |
26.0% |
tool-use-orchestration |
11 |
22.0% |
fine-tuning-adaptation |
10 |
20.0% |
observability-tracing |
10 |
20.0% |
latency-optimization |
9 |
18.0% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
agent-memory + context-engineering + eval-driven-development + human-in-the-loop-gate + model-routing |
1 |
2.0% |
agent-memory + cost-aware-routing + eval-driven-development + fine-tuning-adaptation + human-in-the-loop-gate |
1 |
2.0% |
agent-memory + eval-driven-development + human-in-the-loop-gate + multi-agent-orchestration + retrieval-ranking |
1 |
2.0% |
agent-memory + eval-driven-development + hybrid-search + observability-tracing + query-rewriting |
1 |
2.0% |
agent-memory + privacy-preserving-processing |
1 |
2.0% |
batch-offline-processing + context-engineering + data-quality-loop + eval-driven-development + structured-output-contract |
1 |
2.0% |
batch-offline-processing + cost-aware-routing + eval-driven-development + fine-tuning-adaptation + rag-retrieval |
1 |
2.0% |
batch-offline-processing + cost-aware-routing + eval-driven-development + human-in-the-loop-gate + llm-as-judge |
1 |
2.0% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| retrieval-grounded assistant + evaluation-governed pipeline + cost/latency optimized serving |
7 |
14.0% |
bert-based-sequence-models-for-contextual-product-recommendations (instacart, 2024)
building-a-scalable-retriever-ranker-architecture-malt-s-journey-with-vector-databases-and-llm-powered-freelancer-matching (malt, 2024)
genai-powered-personalized-homepage-carousels-for-food-delivery (doordash, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
4 |
8.0% |
ai-powered-real-estate-transaction-newsworthiness-detection-system (the_globe_and_mail, year n/a)
building-an-ai-interview-coach-for-product-discovery-training (product_talk, 2024)
charlotte-ai-agentic-ai-for-cloud-detection-and-response (crowdstrike, 2025) |
| retrieval-grounded assistant + domain-adapted model pipeline + cost/latency optimized serving |
4 |
8.0% |
ai-powered-semantic-job-search-at-scale (linkedin, 2025)
building-robust-enterprise-search-with-llms-and-traditional-ir (glean, 2023)
jude-large-scale-llm-based-embedding-generation-for-job-recommendations (linkedin, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline |
3 |
6.0% |
ai-powered-incident-response-system-with-multi-agent-investigation (incident.io, 2025)
mercury-agentic-ai-platform-for-llm-powered-recommendation-systems (ebay, 2025)
scaling-an-autonomous-ai-customer-support-agent-from-demo-to-production (intercom, 2023) |
| evaluation-governed pipeline + context/memory system + cost/latency optimized serving |
2 |
4.0% |
personalized-music-recommendation-at-scale-using-llms-and-user-embeddings (spotify, 2026)
real-time-question-answering-system-with-two-stage-llm-architecture-for-sales-content-recommendations (microsoft, 2024) |
| evaluation-governed pipeline + durable async execution + context/memory system |
2 |
4.0% |
ai-powered-performance-optimization-system-for-go-code (uber, 2024)
llm-powered-style-compatibility-labeling-pipeline-for-e-commerce-catalog-curation (wayfair, 2025) |
| guarded enterprise assistant + durable async execution + cost/latency optimized serving |
2 |
4.0% |
genai-powered-accessory-recommendations-for-large-scale-e-commerce-catalog (target, 2025)
scaling-llm-and-ml-models-to-300m-monthly-requests-with-self-hosting (storygraph, 2024) |
| retrieval-grounded assistant + context/memory system + cost/latency optimized serving |
2 |
4.0% |
bridging-behavioral-silos-in-multi-vertical-recommendations-with-llms (doordash, 2025)
context-aware-item-recommendations-using-hybrid-llm-and-embedding-based-retrieval (doordash, 2025) |
Decisions and Rejected Alternatives
Extracted alternatives: 181 (3.62 per raw project). Tradeoffs: 165 (3.30 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Only short-term real-time signals or only long-term batch profiles |
Zillow uses both because batch pipelines capture extended preference windows while real-time behavior catches fast intent changes. |
article / stated |
ai-driven-user-memory-system-for-dynamic-real-estate-personalization (zillow, 2025) |
| Rejected alternative |
Only model-inferred preferences |
Zillow keeps raw interaction history because some experiences need exact replay of what happened. |
article / implied |
ai-driven-user-memory-system-for-dynamic-real-estate-personalization (zillow, 2025) |
| Rejected alternative |
Personalization without explicit privacy controls |
The source says tailored experiences should not come at the expense of privacy, control, or data rights. |
article / stated |
ai-driven-user-memory-system-for-dynamic-real-estate-personalization (zillow, 2025) |
| Rejected alternative |
Big-ship product planning before user validation |
The jet-ski model tests theses quickly and cheaply, then invests big only after evidence appears. |
transcript / stated |
ai-powered-conversational-food-ordering-with-ilo (ifood, 2026) |
| Tradeoff |
batch jobs for slow-moving profiles plus streaming pipelines for transient behavior over one latency profile for all personalization signals |
Daily batch processing can handle complex long-term modeling efficiently, while streaming is needed for rapid shifts in short-term behavior. |
article / stated |
ai-driven-user-memory-system-for-dynamic-real-estate-personalization (zillow, 2025) |
| Tradeoff |
a sufficiently good, concise recommendation over the best fully detailed recommendation |
Voice latency and attention constraints make perfect-but-slow recommendations less usable than fast, clear suggestions. |
transcript / stated |
ai-powered-conversational-food-ordering-with-ilo (ifood, 2026) |
| Tradeoff |
explore handoff or orchestration over ignoring customer-problem reports |
If the bot fails to respond to support complaints, users may perceive the whole assistant as unintelligent. |
transcript / implied |
ai-powered-conversational-food-ordering-with-ilo (ifood, 2026) |
| Tradeoff |
wait until the incident has enough information over force an investigation from a vague incident stub |
Gating avoids positivity-bias output and keeps investigation quality higher when responders most need trustworthy help. |
transcript / stated |
ai-powered-incident-response-system-with-multi-agent-investigation (incident.io, 2025) |
Content generation, localization, and creative operations
Deduped projects: 40. Industries: 8. Provenance: 38 source-backed / 2 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
human-in-the-loop-gate |
24 |
60.0% |
eval-driven-development |
20 |
50.0% |
structured-output-contract |
11 |
27.5% |
tool-use-orchestration |
11 |
27.5% |
guardrails-io-filtering |
10 |
25.0% |
data-quality-loop |
8 |
20.0% |
fine-tuning-adaptation |
8 |
20.0% |
prompt-chaining |
7 |
17.5% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
agent-memory + durable-execution + observability-tracing + tool-use-orchestration |
1 |
2.5% |
async-message-orchestration + guardrails-io-filtering + multi-agent-orchestration + prompt-versioning + rag-retrieval |
1 |
2.5% |
batch-offline-processing + context-compaction + data-quality-loop + eval-driven-development + human-in-the-loop-gate |
1 |
2.5% |
batch-offline-processing + data-quality-loop + human-in-the-loop-gate + latency-optimization + tool-use-orchestration |
1 |
2.5% |
batch-offline-processing + guardrails-io-filtering + human-in-the-loop-gate + privacy-preserving-processing |
1 |
2.5% |
canary-rollout + cost-aware-routing + data-quality-loop + eval-driven-development + human-in-the-loop-gate |
1 |
2.5% |
canary-rollout + cost-aware-routing + guardrails-io-filtering + observability-tracing |
1 |
2.5% |
confidence-thresholding + cost-aware-routing + eval-driven-development + model-routing + multi-agent-orchestration |
1 |
2.5% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving |
5 |
12.5% |
ai-powered-hyper-personalized-email-campaign-automation (promptlayer, 2025)
building-and-managing-production-agents-with-testing-and-evaluation-infrastructure (nearpod, 2023)
building-multilingual-ai-agents-with-translation-pipelines (boundary, 2026) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
4 |
10.0% |
evolution-from-centralized-to-federated-generative-ai-governance (pictet_am, 2024)
from-sms-to-ai-lessons-from-5-years-of-chatbot-development-for-social-impact (one, 2024)
llm-powered-real-estate-search-and-agent-matching (zillow, 2024) |
| evaluation-governed pipeline + guarded enterprise assistant + domain-adapted model pipeline |
4 |
10.0% |
ai-powered-marketing-compliance-automation-system (remitly, year n/a)
building-fair-housing-guardrails-for-real-estate-llms-zillow-s-multi-strategy-approach-to-preventing-discrimination (zillow, 2024)
expert-in-the-loop-generative-ai-for-creative-content-at-scale (stitch_fix, 2023) |
| guarded enterprise assistant + context/memory system + cost/latency optimized serving |
3 |
7.5% |
ai-assisted-product-attribute-extraction-for-e-commerce-content-creation (zalando, 2024)
automated-sports-commentary-generation-using-llms (wsc_sport, 2023)
automating-translation-workflows-with-llms-for-post-editing-and-transcreation (transperfect, 2025) |
| retrieval-grounded assistant + multi-agent/tool workflow + guarded enterprise assistant |
3 |
7.5% |
ai-enhanced-body-camera-and-digital-evidence-management-in-law-enforcement (an_garda_siochanna, 2023)
enterprise-ai-transformation-holiday-extras-chatgpt-enterprise-implementation-case-study (holiday_extras, 2024)
enterprise-wide-generative-ai-implementation-for-marketing-content-generation-and-translation (bosch, 2023) |
| evaluation-governed pipeline + guarded enterprise assistant + context/memory system |
2 |
5.0% |
automated-synopsis-generation-pipeline-with-human-in-the-loop-quality-control (netflix, 2023)
evaluating-product-image-integrity-in-ai-generated-advertising-content (microsoft, 2024) |
| multi-agent/tool workflow + evaluation-governed pipeline + domain-adapted model pipeline |
2 |
5.0% |
ai-powered-customer-segmentation-with-natural-language-interface (klaviyo, 2024)
ai-powered-marketing-content-generation-and-compliance-platform-at-scale (volkswagen, 2025) |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
2 |
5.0% |
automating-search-engine-marketing-ad-generation-with-multi-stage-llm-pipeline (thumbtack, 2024)
global-news-organization-s-ai-powered-content-production-and-verification-system (reuters, 2023) |
Decisions and Rejected Alternatives
Extracted alternatives: 142 (3.55 per raw project). Tradeoffs: 118 (2.95 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Manual-only copywriter QA |
Manual enrichment consumed a large share of production time and left opportunities to reduce errors earlier. |
article / stated |
ai-assisted-product-attribute-extraction-for-e-commerce-content-creation (zalando, 2024) |
| Rejected alternative |
Use raw master-data codes directly with the model |
The model needed human-readable values, then API-compatible codes on the way back. |
article / stated |
ai-assisted-product-attribute-extraction-for-e-commerce-content-creation (zalando, 2024) |
| Rejected alternative |
Stay on GPT-4 Turbo for production generation |
GPT-4o produced better results, faster response times, and lower costs. |
article / implied |
ai-assisted-product-attribute-extraction-for-e-commerce-content-creation (zalando, 2024) |
| Rejected alternative |
Deploying broad AI evidence search as soon as the technology works |
The team cannot use those capabilities until AI legislation allows them, so the system emphasizes evidence infrastructure and human decisions first. |
transcript / stated |
ai-enhanced-body-camera-and-digital-evidence-management-in-law-enforcement (an_garda_siochanna, 2023) |
| Tradeoff |
GPT-4o after comparison over staying on GPT-4 Turbo |
The newer model improved speed and results, but selection was based on domain expert inspection and accuracy tradeoffs. |
article / stated |
ai-assisted-product-attribute-extraction-for-e-commerce-content-creation (zalando, 2024) |
| Tradeoff |
human-first review over fully automatic enforcement decisions |
The organization uses technology to narrow and prepare work, but keeps accountable humans in the decision path for sensitive policing outcomes. |
transcript / stated |
ai-enhanced-body-camera-and-digital-evidence-management-in-law-enforcement (an_garda_siochanna, 2023) |
| Tradeoff |
review/approve/refine plus optional auto-publish over mandatory manual publishing or mandatory full automation |
Guides can stay hands-on for quality control or opt into set-and-forget automation when speed matters more. |
article / stated |
ai-generated-trip-reports-for-outdoor-recreation-guides (guidesly, 2026) |
| Tradeoff |
AI co-pilot with seller editing control over fully automatic ad publishing |
Seller control preserves personal tone and marketplace authenticity while still reducing blank-page friction. |
article / stated |
ai-powered-ad-description-generation-for-classifieds-platform (leboncoin, 2025) |
Enterprise AI platform adoption and governance
Deduped projects: 33. Industries: 11. Provenance: 33 source-backed / 0 summary-only.
Adopted Building Blocks
| Building block |
Deduped count |
Adoption rate within class |
observability-tracing |
17 |
51.5% |
tool-use-orchestration |
15 |
45.5% |
rag-retrieval |
14 |
42.4% |
privacy-preserving-processing |
13 |
39.4% |
human-in-the-loop-gate |
12 |
36.4% |
eval-driven-development |
11 |
33.3% |
safety-policy-enforcement |
11 |
33.3% |
model-routing |
9 |
27.3% |
Common Pattern Combinations
| Pattern combination |
Deduped count |
Share |
agent-memory + eval-driven-development + human-in-the-loop-gate + multi-agent-orchestration + safety-policy-enforcement |
1 |
3.0% |
batch-offline-processing + cost-aware-routing + data-quality-loop + observability-tracing + privacy-preserving-processing |
1 |
3.0% |
batch-offline-processing + data-quality-loop + eval-driven-development + safety-policy-enforcement + tool-use-orchestration |
1 |
3.0% |
batch-offline-processing + hybrid-search + rag-retrieval + tool-use-orchestration |
1 |
3.0% |
canary-rollout + cost-aware-routing + eval-driven-development + model-routing + retrieval-ranking |
1 |
3.0% |
context-engineering + eval-driven-development + human-in-the-loop-gate + model-routing + rag-retrieval |
1 |
3.0% |
context-engineering + eval-driven-development + human-in-the-loop-gate + multi-agent-orchestration + observability-tracing |
1 |
3.0% |
context-engineering + eval-driven-development + human-in-the-loop-gate + prompt-versioning + rag-retrieval |
1 |
3.0% |
Solution Shapes
| Solution shape |
Deduped count |
Share |
Representative projects |
| multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant |
7 |
21.2% |
building-an-enterprise-ai-productivity-platform-from-slack-bot-to-integrated-ai-workforce (toqan, 2025)
company-wide-ai-integration-from-experimentation-to-production-at-scale (trivago, 2025)
grassroots-ai-skills-marketplace-scaling-ai-capabilities-through-bottom-up-engineering (uber, 2026) |
| evaluation-governed pipeline + guarded enterprise assistant + cost/latency optimized serving |
3 |
9.1% |
building-a-multi-provider-genai-gateway-for-enterprise-scale-llm-access (grab, 2025)
building-an-internal-chatgpt-for-enterprise-from-failed-support-bot-to-company-wide-ai-tool (grab, 2025)
building-an-internal-chatgpt-like-tool-for-enterprise-wide-ai-access (grab, 2025) |
| retrieval-grounded assistant + evaluation-governed pipeline + cost/latency optimized serving |
2 |
6.1% |
enterprise-llm-playground-development-for-internal-ai-experimentation (thomson_reuters, 2023)
scaling-ai-assistants-across-swedish-government-offices-through-rapid-experimentation-and-business-led-innovation (government_of_sweden, 2024) |
| retrieval-grounded assistant + evaluation-governed pipeline + guarded enterprise assistant |
2 |
6.1% |
enterprise-genai-virtual-assistant-for-operations-and-underwriting-knowledge-access (radian, year n/a)
genai-transformation-of-manufacturing-and-supply-chain-operations (jabil, 2024) |
| retrieval-grounded assistant + guarded enterprise assistant + context/memory system |
2 |
6.1% |
data-and-ai-governance-integration-in-enterprise-genai-adoption (various, 2024)
leveraging-rag-and-llms-for-esg-data-intelligence-platform (esgpedia, 2024) |
| evaluation-governed pipeline + guarded enterprise assistant + domain-adapted model pipeline |
1 |
3.0% |
federal-government-ai-platform-adoption-and-scalability-initiatives (various, 2023) |
| evaluation-governed pipeline + guarded enterprise assistant + durable async execution |
1 |
3.0% |
large-scale-enterprise-data-platform-migration-using-ai-and-generative-ai-automation (commbank, 2024) |
| guarded enterprise assistant + domain-adapted model pipeline + context/memory system |
1 |
3.0% |
llm-powered-crisis-counselor-training-and-conversation-simulation (crisis_text_line, 2024) |
Decisions and Rejected Alternatives
Extracted alternatives: 123 (3.73 per raw project). Tradeoffs: 108 (3.27 per raw project).
| Type |
Decision or approach |
Rationale |
Evidence provenance |
Project |
| Rejected alternative |
Manual support-ticket classification |
Manual classification was time-consuming, error-prone, fatiguing, and too slow for broad customer feedback analysis. |
article / stated |
ai-powered-customer-feedback-analysis-at-scale (github, 2024) |
| Rejected alternative |
Market AI analytics solutions |
The team needed strict security/privacy alignment and GitHub-specific business metrics for product-team storytelling. |
article / stated |
ai-powered-customer-feedback-analysis-at-scale (github, 2024) |
| Rejected alternative |
Training models on support-ticket feedback |
GitHub applied pre-trained models to analyze ticket text rather than training on customer feedback. |
article / stated |
ai-powered-customer-feedback-analysis-at-scale (github, 2024) |
| Rejected alternative |
Keeping the initial ADX dashboard as the product surface |
Internal feedback showed they needed a tailored tool with business-specific context, advanced filters, and links to internal systems. |
article / stated |
ai-powered-customer-feedback-analysis-at-scale (github, 2024) |
| Tradeoff |
apply pre-trained models without training on ticket feedback over training models on customer support tickets |
This preserves privacy posture while still extracting themes and summaries from feedback text. |
article / stated |
ai-powered-customer-feedback-analysis-at-scale (github, 2024) |
| Tradeoff |
a tailored internal web application with business-specific context over generic dashboards from the first trial |
Product teams needed stories such as top pain points for a specific product/feature, not only generic visualizations. |
article / stated |
ai-powered-customer-feedback-analysis-at-scale (github, 2024) |
| Tradeoff |
a golden path with enablement teams over letting every business unit build arbitrary AI/MCP infrastructure |
The platform makes the right thing easy and the wrong thing difficult while still allowing team-level use cases. |
transcript / stated |
ai-powered-developer-productivity-platform-with-mcp-servers-and-agent-based-automation (bloomberg, 2025) |
| Tradeoff |
MCP hub and inner-source/guild collaboration over parallel duplicate implementations by separate teams |
Discovery and community programs let teams partner, share learning, and deduplicate effort. |
transcript / stated |
ai-powered-developer-productivity-platform-with-mcp-servers-and-agent-based-automation (bloomberg, 2025) |
Interpretation Notes
- Pattern combinations use reviewer-confirmed normalized aliases; unresolved names stay visible in the vocabulary report.
- Solution shapes are compressed from pattern families plus architecture text. They measure composition diversity and do not replace pattern pages.
- If reviewer sampling finds these shape signatures too coarse, the schema still contains the underlying components, alternatives, and tradeoffs needed to refine them.