An AI coach for one athlete.

Training an engineer with a five-day work week toward Ironman 70.3 at the Kinneret in October 2026, then a full Ironman at Klagenfurt in June 2027 — without getting hurt.

Next race — —
—days
—hrs
—min
—sec

Counting down.

Why it exists

Two priorities, in order.

Everything this system does resolves to one of these. When they conflict, the first one wins — not as a slogan, as a rule the code actually follows.

First Stay healthy.

A finish line you reach injured is worse than one you reach slow. Missing a session costs a session. Missing a season costs a season — and at some point, the whole ambition. So the coach is built to hold back a fit, motivated athlete on the days that call for it.

Second Get the best result real life allows.

Not the best result a professional could get. The best one available to someone with a job, a commute, and a family — where the honest training week is the constraint, not a number to feel guilty about. Plan against the week that exists.

Where it stands

Built so far.

  • Complete Intake. History, constraints, thresholds, the full race calendar with A/B/C priorities, and the athlete's own rules.
  • Complete Safety layer. Hard stops, heat protocol, injury red flags and RED-S watch, ahead of any planning logic.
  • Complete Data layer & automation. Readiness now computes daily from real sleep/HRV/RHR, and scheduled runs have delivered every morning for over a week straight — a track record, not a plan.
  • In progress The monitor. A personal daily dashboard now exists — race countdown, today's sessions, open decisions — and updates from real data. Training load (TSS/CTL/ATL/TSB) isn't charted there yet, and the weekly-review write-back is still being tightened.
  • Next Visualizing load. Fitness, fatigue and form as something the athlete can actually look at, rather than something the coach reasons about privately.
The season →