Enterprise-Scale Data Product AI Agent for Multi-Domain Knowledge Discovery

bosch · Automotive · 2026 · source Agentic: yes · promotional bias: low · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enterprise data-product discovery and semantic AI

Bosch needed employees to ask natural-language questions over a huge, fragmented industrial data ecosystem where data lives across divisions, systems, catalogs, ontologies, and domain-specific dialects that generic chatbots cannot interpret.

Constraints:

Bosch has more than 400,000 associates, operates in more than 60 countries, and has a three-digit number of plants.
Every division has a different data ecosystem, architecture, and dialect.
Generic AI does not know Bosch context, such as PT meaning Power Tools instead of Portugal.
The same material number can mean different things in different divisions.
The agent needs semantic access to data products, ontologies, catalogs, LLM knowledge, Bosch internet, and external context.

Architecture

Bosch's DPAI combines Rat Mesh, the Robert Bosch enterprise data mesh; a data marketplace for governance and semantics; and a data product AI agent built on Google Cloud. DPAI uses data-product repositories, ontologies, enterprise data catalogs, LLM farms, Bosch context, internet/internal documents, and APIs/RAG-like access so users can ask natural-language questions over semantically described data products.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  rat-mesh-data-mesh["Rat Mesh data mesh"] --> data-marketplace["Data marketplace"]
  data-marketplace --> data-product-repository["Data product repository"]
  data-marketplace --> dpai-agent["DPAI agent"]
  data-product-repository --> dpai-agent["DPAI agent"]
  ontologies["Ontologies"] --> dpai-agent["DPAI agent"]
  enterprise-data-catalog["Enterprise data catalog"] --> dpai-agent["DPAI agent"]
  llm-farm["LLM farm"] --> dpai-agent["DPAI agent"]
  google-cloud-infrastructure["Google Cloud infrastructure"] --> dpai-agent["DPAI agent"]
  dpai-agent --> business-users["Business users"]

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Rat Mesh data mesh`	Enterprise data mesh that structures Bosch data across functional domains.	`Data marketplace`
`Data marketplace`	Governance and semantic layer for discovering and assembling data products.	`Data product repository`, `DPAI agent`
`Data product repository`	Stores descriptions, LSAs, access information, and data-product metadata.	`DPAI agent`
`Ontologies`	Represent Bosch domains, divisions, plants, machinery, and products.	`DPAI agent`
`Enterprise data catalog`	Provides technical metadata from Informatica EDC and Unity Catalog.	`DPAI agent`
`LLM farm`	Provides models from GPT, Gemini, and other providers.	`DPAI agent`
`DPAI agent`	Middleware/data-product AI agent for natural-language communication with Bosch data.	`Google Cloud infrastructure`, `Business users`
`Google Cloud infrastructure`	Provides infrastructure for agents to connect to data.	`DPAI agent`
`Business users`	Ask natural-language questions such as revenue or OEE by plant/division.	—

Data flow

Data products are structured in Rat Mesh and governed in the data marketplace; metadata, access, ontology, catalog, and Bosch context feed DPAI; users ask questions in natural language; DPAI resolves Bosch-specific meaning and division context, connects to data through APIs/RAG, and returns fast, correct business answers.

“The first component is to give structure to our data across the whole organization across all the functional domains.”transcript, stated

“DPAI combines the foundational levels from WVA. He was speaking about the data sources, data governance, Bosch context, data marketplace.”transcript, stated

Why this architecture

Bosch built DPAI on top of data mesh, marketplace, ontologies, catalogs, and Bosch-specific context because a generic chatbot cannot answer enterprise questions without system access and semantic understanding of Bosch terminology.

“our AI doesn't have access to our systems and it doesn't know the Bosch context.”transcript, stated

“we have to teach the AI also the Bosch language and the different dialects within Bosch.”transcript, stated

Models used: DPAI · Google Cloud · GPT · Gemini · Informatica EDC · Unity Catalog · Rat Mesh · Data marketplace · RAG · APIs

Scale

Bosch has more than 400,000 associates worldwide.
“more than 400,000 associates worldwide”transcript, stated
Bosch operates in more than 60 countries.
“operating in more than 60 countries”transcript, stated
Bosch has a three-digit number of plants.
“having a tripledigit number of plants.”transcript, stated
Plant 103 and OEE are used as example manufacturing contexts DPAI must resolve.
“DPAI knows that the plant 103 belongs to the business division PT”transcript, stated
Bosch currently has multiple ontologies per domain and business division, with a goal of one connecting ontology.
“At the moment we have multiple ontologies for every domain and very every business division”transcript, stated

Alternatives considered (roads not taken)

Ask a generic internal chatbot without data-system access.

Rejected because: Ask Bosch could not answer because it lacked commercial-data access and Bosch context.

“I don't know. I don't have access to our um commercial data. Please ask your controller.”transcript, stated

Treat division acronyms and identifiers as generic language.

Rejected because: PT could be misread as Portugal rather than Bosch's Power Tools division.

“PT would probably have been interpreted as Portugal. In fact, it's one of our divisions. It's power tools”transcript, stated

Rely on data mesh alone.

Rejected because: Bosch adds marketplace semantics, data product repository, ontology, catalog, and LLM context on top.

“Second component is our data marketplace to bring in governance to bring in the semantics.”transcript, stated

Patterns

`rag-retrieval` (retrieval)

DPAI uses RAG-like and API access to connect agents to Bosch data.

“Yeah, we are using rack. So the whole DPIA is built on Google cloud”transcript, stated

`tool-use-orchestration` (orchestration)

The agent accesses data through APIs and Google Cloud infrastructure rather than relying only on model memory.

“via APIs we are accessing the data”transcript, stated

`hybrid-search` (retrieval)

DPAI combines marketplace metadata, catalog metadata, ontologies, LLM knowledge, internal documents, and external context.

“you have context in the world that order of Bosch. So news gardener reports and other documents. But you have also Bosch internet.”transcript, stated

`structured-output-contract` (data)

Data products carry descriptions, access details, LSAs, and semantic metadata so the agent can reason over them.

“you have their description, you have LSAs, you know how to access the data and who is allowed to access this data.”transcript, stated

Tradeoffs

Build semantic foundations before the AI agent.

Chose Data mesh, marketplace, data product repository, ontology, catalog, and context layers. over Connecting a chatbot directly to raw data systems..

Rationale: The agent needs meaning, access, and Bosch language to answer correctly.

“So give data a meaning and the third component is the DPAI our data product AI agent.”transcript, stated

Use DPAI as middleware.

Chose A connected agent layer that can be embedded or connected A-to-A. over A standalone chatbot disconnected from Bosch data products..

Rationale: DPAI must connect to multiple enterprise systems and data products.

“DPAI it's kind of middleware so it's connected so it can be connected by A to A connection or deeply embedded”transcript, stated

Represent domain knowledge with ontologies.

Chose Domain/division ontologies moving toward one connecting ontology. over Text-only document retrieval..

Rationale: OEE, plants, products, divisions, machinery, and material numbers require structured semantic context.

“one ontology that connects business divisions, machineries, plans and products.”transcript, stated

Gotchas (what bit them)

Enterprise acronyms and identifiers are ambiguous without business context.

Root cause: A generic model could interpret PT as Portugal rather than Power Tools.

Resolution: Encode Bosch-specific context, ontologies, and division semantics into DPAI.

“Like PT would probably have been interpreted as Portugal.”transcript, stated

The same identifier can have different meanings by division.

Root cause: Bosch divisions use different data ecosystems and dialects.

Resolution: Use semantic models and domain ontologies that connect divisions, plants, machinery, and products.

“the same material number can have a very different meaning in the different divisions”transcript, stated

A data mesh pilot can outgrow its initial scalability assumptions.

Root cause: The pilot began from one business-division requirement before spreading.

Resolution: Add enterprise data marketplace and semantic layers to scale data products.

“we started very small. We started without thinking about an enterprise scalability”transcript, stated

Evaluation

The presentation evaluates DPAI through example questions showing that generic chat fails without data access/context, while DPAI should provide fast and correct answers grounded in Bosch data marketplace, ontologies, catalogs, and APIs.

Metrics: Fast and correct answers · Semantic resolution of Bosch-specific terms · Data product access and permissions · Coverage of divisions/domains · Plant/OEE question handling

“At the end, we have a platform that gives you fast and correct answers.”transcript, stated

“The normal standard chatbot accesses PT or ABC data in this case sales data and gives correct answer.”transcript, stated

Outcomes

Bosch presents DPAI as an in-development data-product AI agent for natural-language communication with Bosch data.
The architecture combines data mesh, data marketplace, data product repository, ontologies, catalogs, LLM farm, and Google Cloud.
The source shows example improvements over a generic chatbot, but does not report production KPIs.

Bosch's operating scale is very large: 400,000+ associates, 60+ countries, and a three-digit number of plants.

“DP AI data product and AI agent that enables natural language communication with Bosch data.”transcript, stated

“I guess probably we have the largest industrial data ecosystem in the world.”transcript, stated