AI Knowledge Base & Internal Assistant

Description

New engineers take about six weeks to become productive, and most of that is learning where things are documented — Confluence, a decade of Slack, README files, and a few people's heads. We want an internal assistant over the written sources. Confluence and repository documentation are in scope. Slack history is not, for this phase: too much of it is informal, contradicted later, or private, and we would rather have a smaller trustworthy corpus than a large unreliable one. It should answer with links into the source rather than a summary alone, so people learn where things live instead of depending on the assistant forever. Where documentation is contradictory or out of date, we want it surfaced rather than smoothed over — that list is useful to us in its own right.

• Index Confluence spaces and repository documentation, respecting existing access controls
• Answer with links into the source material, not summaries that hide where the answer came from
• Flag contradictory or stale documentation as a report we can act on
• Provide a feedback control on each answer, and a weekly view of what it got wrong
• Keep the index current as pages and READMEs change

• Internal assistant deployed behind our SSO
• Indexing pipeline for Confluence and repository docs
• Stale and contradictory documentation report
• Feedback capture with a weekly summary for the platform team