AI · LLM · computer vision · PWA · React · Go
Bilberry
A diet assistant that learns you: it counts calories from a photo of the meal, derives your real metabolism from the weight trend, and plans a week of home cooking - taking your lab results into account.
bilberry.health → (project site)The challenge
Calorie tracking dies from friction: searching products, weighing, typing. And a calorie target from an online calculator is an average for somebody else - not for the person using it.
The solution
A meal is logged with a photo or a single sentence - the model recognises the ingredients and computes the macros; the user only confirms. The optimiser lines the weight trend up against what was actually eaten and adjusts the calorie target weekly - instead of an internet formula. On top of that: a weekly plan of home-cooked meals with a shopping list that subtracts what the pantry already holds, and lab results from a PDF, where an out-of-range marker changes what lands on the plate.
Scope & technology
The whole product: React + TypeScript, a Go API, PostgreSQL, LLM integrations (meal recognition from a photo, an assistant with access to the diary, plans, and labs), a barcode scanner, an offline-capable PWA, an Android app, light and dark modes, PL/EN. Health data is isolated at the database level, and an account can be deleted in one click.