AI product2026In-house product
Annam
Photograph a plate of Indian food, get calories and macros back per item — idli, sambar, poha, the katori of dal included.
- Our part
- Product, design, engineering
- Time
- 3 weeks
- Built with
- Next.jsVision modelsVercel
Where it started.
Every calorie app on the market is trained on Western plates. Photograph a South Indian thali and it returns nothing useful, so anyone eating normal Indian food gives up in a week.
What we did.
A vision model tuned for Indian plates, where the hard part is not detection but portions: a katori, a ladle of sambar, a spoon of ghee. Built end to end, from the camera screen to the macro breakdown.
- 01Wrote the vision prompt around Indian portion language — katori, ladle, spoon of ghee — rather than grams nobody weighs
- 02Built the whole path in one screen: camera, upload, per-item macros, correction
- 03Normalised model output defensively, so a bad response degrades instead of crashing the page
- 04Kept a second provider behind the first, because a single vision API is a single point of failure
Where it stands
Built and running on preview. It is waiting on a production vision key before it goes public — the honest status, rather than a launch date we have not hit.
What this took.
Next
Reaching Dreams
Online store
Want one of these?
- A scope and a fixed price before anything starts
- A live link from the first week
- The code in your account at the end