DSF

I built Cahier for myself first. I'm one of the 83% of European adults drowning in contradictory health information. The wellness industry sold me solutions; they failed; I concluded I was broken. Then I read the literature — the actual one — and the principles worked. So I built the tool I'd needed. Cahier is a calm-tech health app: one concept a day from a primary-literature corpus of 310 papers across nutrition, training, cardio, sleep, and recovery — curated to 48 for the runtime RAG. Habit psychology shapes the cadence (small daily exposure beats 12-week transformations); Self-Determination Theory shapes the voice (17 banned fitness-bro words); calm technology (Weiser) shapes the surface (hand-painted, biomorphic, no streaks); trust epistemics shapes the chat (every reply passes a citation validator that strips any [^N] not actually retrieved above threshold). Powered by Claude Opus 4.7 with adaptive thinking and 1M context, hybrid RAG (server-side embedding → on-device cosine top-k), and a hand-rolled SSE transport — no SDK. The build held itself to the same evidence standard. compound-agent (github.com/Nathandela/compound-agent), an open-source harness I wrote, drives each task through a test-first cycle with multi-model review gates: Opus implements, fresh Sonnet and Opus review; failures become captured lessons. 70k+ lines of TypeScript shipped across multiple 5-to-15 hour autonomous Claude Code sessions on my laptop while I worked or slept. The thesis: adults aren't broken. Given honest evidence, plain language, and the daily cadence of habit, people work themselves out.