So, I built a thing for myself that I'd like to share with others. The common pattern for AI health tools is one general chatbot answering questions with no real context. I went the other way and built eight specialist agents, training, nutrition, longevity, recovery, biohacking, running, and so on, that all read from one shared memory of the full history and synced wearable data.
The interesting design problem was that a search box that knows nothing about you can only give the average answer. Routing a question to the right specialist while keeping one shared context made the answers specific to the actual person. There is also an n of 1 experiment mode that tracks an intervention window and reports whether the metric moved.
It is early and solo. Free while I learn. Happy to talk architecture, the model routing and the shared memory design, in the comments. I will keep the link there too.
[link] [comments]