← Back to projects
PERSONAL PRODUCT · OFFLINE-FIRST PWA

Fede Training

A personal training PWA I build for myself and actually use: structured programming, one exercise at a time, and a nutrition plan that shows its own numbers instead of asking me to trust them.

Four or five times a week, in a gym in Dublin, this is the thing in my hand between sets.

It is not a product with users. It is the place where I keep a strength programme, a calisthenics ladder and a nutrition plan honest: the exercise decides which fields appear, a recorded session is never rewritten by a later change, and the numbers on screen are computed from what is actually in the plan rather than asserted.

The interesting part is not the app. It is what maintaining it forces: every change is checked against 657 automated tests and a reviewed reference of every screen, because the data underneath is a year of my own training that cannot be regenerated if I break it.

Three screens of the app at phone size: a training day, a single exercise, and the progress view

Why I built it & impact

Generic fitness apps hand you a wall of fields and trust you to fill them in. This one decides what to ask: a hold asks for seconds and no kilos, a unilateral exercise is entered once with the asymmetry read back later, and a session opens on a single exercise rather than a column of seven cards. Every change is driven by a problem I hit while training, not by a feature list.

Training

Structured strength, hypertrophy and calisthenics programming, including Pull Ups, L Sit, Frog Stand and Hanging Knee Raises.

Progression

Previous-session data sits directly within each exercise, showing sets, reps and weight or time where available, so progression can be compared while training.

Training memory

Workout history gives the Coach context about previous sessions and progression, turning past training data into useful context rather than a static archive.

AI Coach

A rules-based assistant edits the plan through one vetted path, so a change can never corrupt recorded history. Model-backed coaching is deliberately deferred until the data underneath it is trustworthy.

Product evolution

Stable exercise IDs, 616 automated tests, and a release reference that every screen is compared against: the programme can evolve without rewriting what was already recorded.

From tracking to contextual coaching

Auditing the nutrition section found three different calorie targets coexisting in the same app, none matching the 2000 kcal my trainer set. The fix was not to invent a new plan but to surface the real numbers, the gap, and the margin of error on the estimate.

Built with

Vanilla JS PWA · offline-first, local storage only · JSON export/import · Cloudflare Pages · 616 tests in CI

Live

Public, no account, no backend. Training data stays on the device.

Make it useful.
Want to build something?

Let’s connect.

Email me  →