Manufacturing Decision SupportCross-Source ReasoningAgent Workflow

Manufacturing Knowledge Assistant

A domain-aware assistant built for analysis rather than generic Q&A.

AI Engineer / Project Leader2023 - 2025AUO Corporation

Decision Assistant · Multi-Agent Architecture

LangGraph Multi-AgentNeo4j Knowledge GraphCross-Source Reasoning
  1. 01

    Fab Question

    Cross-Source Issue Query

  2. 02

    SOP Retrieval

    Process Docs & RAG

  3. 03

    SQL Tooling

    Live Yield & Chamber Data

  4. Domain Graph04

    Neo4j KG

    Equipment–Process Links

  5. 05

    Agent Routing

    Multi-Step Tool Orchestration

  6. 06

    Decision Aid

    Grounded Root-Cause Path

Evolved from basic text search into a LangGraph workflow coordinating RAG, SQL, and Neo4j process graphs.

RAG + SQL + Neo4j · Investigation Flow
Workflow overview. Some names and data are anonymized or recreated for confidentiality.

Before

Engineers had to piece together manufacturing issues across documents, tables, and process knowledge before they could even form a useful hypothesis.

After

Retrieval, SQL tools, knowledge modeling, and agent workflow design were combined into a decision-support system that could help structure investigation work.

Context & Background

  • Engineering teams needed help investigating manufacturing issues that crossed documents, data tables, and process knowledge.
  • The challenge was less about text generation and more about supporting real analytical judgment.

Constraints

  • Relevant information lived across multiple systems and required domain context to interpret.
  • Any AI support had to be structured enough to avoid loose, ungrounded answers.

Problem Solved

  • Manufacturing investigations required jumping across multiple sources before a useful analysis could begin.
  • Basic retrieval alone was not enough for operational decision support.

My Role

  • The assistant was reframed from a retrieval tool into a decision-support workflow.
  • Tool use, reasoning steps, and domain representation were designed to work together as one system.

System Design

  • Combined Azure OpenAI models, retrieval, SQL access, domain relationships, and LangGraph agent orchestration into one structured analysis flow.
  • Used knowledge representation and tool routing so the assistant could support investigation instead of only retrieving text.

Implementation Highlights

  • Expanded the assistant from document retrieval to structured decision-support workflow.
  • Integrated Azure OpenAI APIs with retrieval, SQL tools, Neo4j knowledge modeling, and LangGraph orchestration.
  • Focused the system on investigation support instead of open-ended chatbot behavior.

Outcome & Impact

  • Made manufacturing investigations easier to structure and faster to start.
  • Moved the assistant closer to practical engineering support rather than generic Q&A.

Tech Stack

PythonAzure OpenAILangGraphRAGNeo4jSQL Tools