Production AI for engineering knowledge

A production chat application that helped users find answers in engineering documents with source references and tenant-aware access boundaries.

Angular
C#
.NET
Azure OpenAI
Azure AI Search
Azure

Problem

Engineering teams needed faster ways to find answers inside large document collections and related database records without losing the context of where an answer came from or which company tenant should see it.

My contribution

I designed and built a production chat application using Azure OpenAI and Azure AI Search with retrieval-augmented generation. The work covered the frontend experience, backend services and the lifecycle steps needed to take the system into production. I designed the tenant data-isolation model so each company tenant stayed within its own access boundaries. Answers returned with five source references so users could verify what the system used.

Technical decisions

I chose Azure OpenAI and Azure AI Search to support retrieval-augmented generation over engineering content. The application worked over 300,000+ engineering documents and millions of database records. Angular, C#, .NET and Azure formed the delivery stack so the same team could own API, UI and deployment concerns together.

Outcome

The application was deployed to 30,000 users across 65 multi-tenant companies. Users could ask questions against engineering knowledge and receive answers with source references inside their tenant boundaries.

Next project

Enterprise search at document scale