AI Assistant Integration for Manufacturing Execution System (MES)

42q · Tech · 2025 · source Agentic: no · promotional bias: med · source-backed · extracted with prompt 0.5.2 (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: Manufacturing execution support

42Q users struggled with a complex MES, frequent updates, terminology differences, and limited access to expert administrators.

Constraints:

MES complexity
multilingual support
live production data
action safety

Architecture

42Q embedded Arthur inside its MES portal as a Bedrock-backed assistant. Phase one loaded product documentation and transcribed training videos so the chatbot could answer how-to questions, remain multilingual and contextual, and cite its sources. Phase two connected Arthur to the live 42Q MES database so users could ask about shop orders, part numbers, routes, defects, and other production state. Guardrails and portal login constrain access while the team continues to evaluate whether Arthur should take direct production actions.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  arthur-portal-chatbot["Arthur portal chatbot"]
  documentation-knowledge-base["documentation knowledge base"]
  amazon-bedrock["Amazon Bedrock"]
  live-mes-database["live MES database"]
  source-citations["source citations"]
  guardrails["guardrails"]
  production-users["production users"]
  production-users --> arthur-portal-chatbot
  arthur-portal-chatbot --> documentation-knowledge-base
  documentation-knowledge-base --> source-citations
  arthur-portal-chatbot --> amazon-bedrock
  amazon-bedrock --> guardrails
  arthur-portal-chatbot --> live-mes-database
  arthur-portal-chatbot --> source-citations

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

Components

Component	Role	Interacts with
`Arthur portal chatbot`	Embedded assistant available after users log in to the 42Q portal.	`documentation knowledge base`, `Amazon Bedrock`, `live MES database`, `source citations`
`documentation knowledge base`	Loaded 42Q documentation and transcribed training videos used for RAG-style MES help.	`Arthur portal chatbot`, `source citations`
`Amazon Bedrock`	AWS generative-AI service used to build Arthur, including guardrails and agent capabilities.	`Arthur portal chatbot`, `guardrails`
`live MES database`	Connected 42Q MES database queried by Arthur for shop orders, part numbers, routes, defects, and related production data.	`Arthur portal chatbot`
`source citations`	Answer references pointing users back to documentation or training-video sources.	`Arthur portal chatbot`, `documentation knowledge base`
`guardrails`	Bedrock/portal safety controls around authenticated access and assistant behavior.	`Amazon Bedrock`, `Arthur portal chatbot`
`production users`	Manufacturing users and managers who ask questions in natural language during production work.	`Arthur portal chatbot`

Data flow

After a user logs into the 42Q portal, Arthur receives a natural-language question. For product/help questions it retrieves from documentation and transcribed training videos and returns a cited answer. For production-data questions it runs live queries against the connected MES database, presents current operational data in natural language, and remains inside the portal/Bedrock guardrail boundary.

4 supporting quotes

“We added all our documentation that we had. We added we transcribed all our training videos where we explain how 42Q works where we give tips and hide uh guidance.”transcript, stated

“Then we embedded the chatboard into our portal so that it is always there. It is once you are locked into 42Q you have it right away.”transcript, stated

“for every answer that he gives, he points you straight to the resources where he found this answer”transcript, stated

“we decided to connect Arurther to our 42qmees database which simply means ather itself runs live queries against the ex the connected database”transcript, stated

Why this architecture

42Q needed an assistant that reduced dependence on scarce MES administrators without forcing users to learn another tool. Embedding Arthur in the authenticated portal put help where users already work, RAG over documentation/training videos made the MES domain available, live database access added current production context, and guardrails kept the system in an answer-and-cite posture before action-taking is approved.

4 supporting quotes

“this is our Amazon Bedrock service which is what 4TQ uses to build their application and this includes various capabilities that are built in the uh the service for instance guard rails and agents”transcript, stated

“It is by design uh multilingual which means he understands almost any language. Uh he stays in context which means whenever you have a question you may not really understood it.”transcript, stated

“We have guardrails in place and it's an out ofthe-box solution.”transcript, stated

“So until now it understood the the software it understood the system but not the data. So we decided to connect Arurther to our 42qmees database”transcript, stated

Models used: Amazon Bedrock foundation models

Scale

Arthur is deployed in the 42Q portal rather than as a standalone demo.
“Then we embedded the chatboard into our portal so that it is always there. It is once you are locked into 42Q you have it right away.”transcript, stated
The database-backed phase exposes production objects such as shop orders, part numbers, routes, and defects.
“it shows you something like shop orders, part numbers, routes, defects, whatever you may very likely want to know.”transcript, stated
A production manager saw night-shift value because non-expert workers could ask about remaining shop orders.
“they were really impressed because they said, "Wow, that fixes one of my issues in night shifts when my production control is not uh around there anymore."”transcript, implied

Alternatives considered (roads not taken)

Documentation-only interactive helper chatbot

Rejected because: It knew the 42Q software but not current production data, so the team added live database queries.

“So until now it understood the the software it understood the system but not the data. So we decided to connect Arurther to our 42qmees database”transcript, stated

Immediate direct production actions

Rejected because: The approved v1 record shows the team is still deliberating action-taking; the v2 source supports an answer/citation/data-query architecture with guardrails first.

“The chatbot has been successfully deployed, with users reporting faster access to information, improved training capabilities, and reduced reliance on expert staff, though the team is still deliberating on whether to allow Arthur to take direct actions in production environments.”summary, stated

Patterns

`rag-retrieval` (support)

Arthur retrieves documentation and transcribed training knowledge to answer MES questions.

“The solution uses RAG (Retrieval Augmented Generation) with Amazon Bedrock, incorporating documentation, transcribed training videos, and live database queries.”summary, stated

`safety-policy-enforcement` (control)

The team is cautious about allowing the assistant to take direct production actions.

“improved training capabilities, and reduced reliance on expert staff, though the team is still deliberating on whether to allow Arthur to take direct actions in production environments.”summary, implied

Tradeoffs

Whether to let the assistant act in production.

Chose answering and citing first over direct production changes immediately.

Rationale: The team prioritizes safe information access while evaluating the risk of action-taking.

Gotchas (what bit them)

Terminology mismatches made MES help harder than multilingual translation alone.

Root cause: Users describe the same manufacturing concept with different words such as product, part number, or material number.

Resolution: Ground answers in MES context and domain vocabulary instead of literal keyword lookup.

“ingual support (42Q offers 7-8 languages), there's the challenge of contextual understanding - different users may refer to the same concept differently (e.g., "product," "part number," "material number")”summary, stated

Evaluation

42Q reports product release and user feedback rather than a public benchmark.

Metrics: faster information access · training support · night-shift usefulness

“dd-on to this one because we have a lot of Q&A sessions. So whenever we have a question now we have Arthur and he gives the answer. Good. So the first feedback that we got because we released it. It's right now in our product. Wow that's pretty cool. We got very fast responses comparing to readin”transcript, stated

Outcomes

Users reported faster answers and reduced reliance on busy MES administrators.

Assistant/RAG system with possible future action-taking; marked non-agentic because current extracted behavior is support/querying.

“When I to when I showed this to one of my production manager, they were really impressed because they said, "Wow, that fixes one of my issues in night shifts when my production control is not uh around there anymore." Then I can just have anyone who have access to the system use other and say, "H”transcript, stated