An agent that answers Revenue Squared members’ questions and routes them to the right expert, saving me over 10 hours a week. It runs on three years of Slack text plus video and audio from the community, anonymised and then structured into a compact index of who knows what, and serves roughly 250 members a week.
A community with hundreds of members accumulates answers faster than anyone can find them. Newcomers asked questions that had been answered three times. The archive was too big to hand to a model on every question and too messy to search by keyword.
The source was three years of Slack text plus video and audio. A one-off pass anonymised it and structured it into a knowledge base by theme. On top of it sits a much smaller index: for each theme, the top experts with credentials and post counts, the best resources, and the canonical threads. A JSON twin serves scripts without parsing text. When a member asks something, the answering agent reads the index first and only opens the full archive when the index cannot answer. A weekly job pulls what changed and updates the index; a quarterly job rebuilds it from scratch.