Agentic AI Assistant for Engineering BOMs

Engineers using a large bill-of-materials platform had to manually search and assemble parts lists across a sprawling catalog. The goal was to let them work in natural language: "find 16-gauge green wire," "compare 4-inch and 6-inch butterfly valves," even create and modify BOMs conversationally, without ever giving an AI unchecked write access to engineering data.

We built a production assistant on a six-tool Azure OpenAI function-calling agent. Read tools (asset search, project lookup, BOM summarization) answer questions over a hybrid RAG index built on Azure AI Search with text-embedding-3-large; write tools (create BOM, add parts) are gated by a confirm-before-write flow where the user explicitly approves any change before it is committed. The chat layer streams responses over SSE with structured error and cancellation handling.

The assistant shipped into production and became the team’s canonical reference implementation for AI features, with its chat layer extracted into a shared component for reuse across other applications. Engineers can now query and assemble BOMs conversationally while every write stays under human control.