Client
A health founder
Sector
Health · community
Services
AI, RAG, full-stack
Role
Sole builder
Status
In production
A retrieval assistant over a founder's body of work — answering his community with citations.
The founder had years of published health content and a growing community asking the same questions repeatedly. He didn't want a generic chatbot — he wanted an assistant grounded in his own material that answers with sources, so the community could trust it and staff weren't fielding the same questions over and over. The goal wasn't scale for its own sake; it was making sure the knowledge he'd already built actually reached the people asking for it, with enough transparency that they could follow the reasoning themselves.
Discovery. Mapped the questions the community asks most — the recurring ones that staff handled by hand — to understand what the assistant needed to be good at.
Corpus. Gathered and indexed the founder's published content: the body of work that should power every answer.
Retrieval. Built a hybrid search over that corpus, so the assistant finds the most relevant material before composing a response.
Answers. Composed responses that cite the sources they used, with evaluation to keep quality honest over time.
Every response draws from the founder's own content and shows its work — so the community gets answers they can verify, not a black box.
Surfaces common starting points so members know what to ask, lowering the barrier to getting useful answers on their first visit.
Deployed and in active use — improving against real questions, not sitting in a demo environment.
Members self-serve confident, sourced answers rather than waiting for staff to respond. The founder's team spends far less time on repeat questions, and the knowledge he'd already built finally has a surface that puts it to work. It runs in production and continues to earn use.
Sole builder: product, retrieval pipeline, and full-stack engineering, end to end. Built with a custom RAG pipeline, hybrid search over the founder's corpus, and streaming responses.