A codebase assistant inside Ford's developer portal
A chat with RAG over the repositories, built alone in two months.
50+
people using it
Context
Support analysts, developers and business people needed answers about how the systems worked. The answers were in the code, spread across many repositories.
What I did
Working alone, I designed and built a chat inside Backstage in two months. It indexes the GitHub repositories with RAG on pgvector, answers questions and uses MCP tools I wrote, such as one that opens tickets. Teams could choose Ford's internal model or a Google model.
Built with: Python, Next.js, RAG with pgvector, MCP, LLMs, Backstage
How it works
Hover, focus or tap a step to see what it does.
Result
Used by 50+ people across support, development and business. It was later replaced by a newer generation of internal tools.