Climb Trainer

A mobile-first climbing training app: pick a target grade, get a generated plan that covers the different aspects of climbing, then follow the session along with a timer and audio cues while it logs itself — and read it all back as a training diary.

• Live: https://climb.kilianmc.com (standalone app)
• Also runs inside https://kilianmc.com as the climbTrainer Module Federation remote — the third showcase project in the portfolio, and the first with a database and a backend that isn't JavaScript.

What it does

1. Target grade in. Log in, then answer four questions: the grade you climb now and the one you are training for, one strength and one weakness, your weekly availability, and anything that is currently injured.
2. Plan out. A pure-Python plan generator turns the gap between your current and target grade into a phased plan — base → strength → power → power endurance → peak/taper, with a deload every fourth week — allocating volume toward your weakest aspects, skipping anything an open injury contraindicates, and naming the gear that would unlock a session it had to build thin.
3. Guided session player. A protocol interpreter compiles any prescription (max hangs, repeaters, 4×4s, EMOM, min-edge, limit boulder, circuits) into one timeline of prepare / work / rest / open phases. Full-viewport colour changes and a huge countdown carry the cues, with synthesized audio and haptics on top. Every tap is a local write first, so it works with no signal in a basement gym.
4. Training diary. A note on the session it describes, plus free-standing entries for how you slept, how your skin is and how it felt — read back as a trend, not a replay.

Stack

Frontend — React 19 · TypeScript 6 (strict) · Vite 8 · SCSS · Vitest + React Testing Library · @module-federation/vite (remote role).

Backend — FastAPI · Python 3.13 · sync SQLAlchemy 2 · psycopg3 · Alembic · pytest · ruff.

Data — Neon Postgres. The plan tree is fully relational, closed vocabularies are native enums, and grades carry an integer ordinal rather than a display string — one contiguous ladder per discipline, so V-scale and Font boulder grades sit on the same rungs while boulder and rope grades are deliberately not treated as comparable.

Deploy — a single Vercel project serves both the SPA and /api/*, so the API is same-origin for the standalone app. uv manages the Python toolchain.