Fuel

what it is

A mobile-first fitness tracker, installable to your phone like an app, for logging food, weight, and workouts. Pick a goal and the Today tab shows your calorie, macro, and fibre rings against what you have eaten. Log food three ways: scan a barcode, search Open Food Facts, or photograph the meal and let a vision model estimate what is on the plate, which is the path that covers home cooking and anything without a label. A Train tab runs a push, pull, and legs split and shows your last session for each lift so you can beat it. It tracks and visualises; you set the targets, not a coach.

what is unique

It is a real product, not a personal script. Full multi-user cloud sync on Supabase, with row-level security so each person only sees their own data, yet it still works completely offline. The photo estimate is treated as a starting point rather than an answer: it lands in an editable entry you confirm before it is saved, because a vision model guessing at portion size is genuinely uncertain and a number that silently saves itself is worse than one you were asked to check. And every target is transparent and editable, not a black box: the calorie and macro numbers use standard sports-science formulas, each shown with its source. That is what took it from a weekend build to something real testers open every day.

how it was built

The whole app is a single 79KB index.html of vanilla JavaScript, no framework and no build step. Supabase was added later as a purely additive backend over an already-working offline app, so sync is a bonus, never a dependency. Nutrition data is free from Open Food Facts. It installs to a phone through a web manifest and service worker. Built almost entirely through conversational development with Claude, by an MBA with no engineering background, which is the point.

gtm angle

Shipping an AI-built product is the easy part. Daily use is the real test, and this is the one build where real people actually open it every day. That is the bar GTM is ultimately judged on, not launch but return, and it is a harder one to clear than getting something live.

from the live build

01 / 04

Four ways in, because friction is why food tracking gets abandoned. Photo, barcode, search, or a saved meal, with the day still empty underneath.

02 / 04

The training half of the same app: the session plan for the day, and sets logged weight by reps as they happen.

03 / 04

Search runs against the Open Food Facts corpus, so real supermarket products come back with real per-100g numbers.

04 / 04

The day once a meal has landed: calories left against target, the four macros underneath, weight and steps below that.