Autopilot pulls together all the knowledge a team spreads across different tools, like product docs in Notion, decisions made in Discord, code in GitHub, and issues in Linear, and keeps it in one place that stays up to date. It also watches for drift, so when those sources start to contradict each other, you find out.
The hard part was extraction and reconciliation: taking messy content from all these sources and turning it into structured claims you can actually trace back to where they came from, then catching conflicts before they quietly rot the docs. So we built it to get extraction quality right first, then added the intent graph on top.
I was the PIC on this, working with one other engineer. I owned the system design and architecture and wrote the code, and I worked with the team on the product side too. Building with AI-assisted tools was a big part of how I got Phase 1 designed, built, and shipped in two sprints.