I spent a year building the fitness app I kept wishing existed. It launched on Product Hunt today.

Every app I trained with was a logbook. It recorded what I did and then went quiet. Nobody answered the only question that matters: given what just happened, what should I do next? A human coach answers that for $200 to $400 a month, so most lifters guess, plateau and quit.

Most of what a good coach does is a rule system, not magic. Progress the load when the reps land. Deload when they don't. Move calories when the rolling weight trend drifts. Rebalance weekly volume when a muscle group falls behind. So I built that as deterministic engines rather than LLM calls, which means it's testable, reproducible and nearly free at the margin. That last part is why the free tier can be genuinely deep instead of a crippled trial.

The stack: React PWA on Vercel, Fastify on Railway, Postgres and Prisma, Supabase for auth, Stripe for billing and coach revenue share. Offline-first with IndexedDB, because gym basements have no signal. Solo build, AI-assisted throughout.

Where I am honestly: 41 users, all organic, first signup two weeks ago, zero revenue by choice while I work on retention. Today is the first real distribution attempt.

Happy to answer anything about building solo at this scope, or about why I refused to put an LLM in the core loop.