Nourish

Most calorie trackers are built around Western food. A typical Indian vegetarian meal is one plate of several dishes, such as dal, sabzi, roti and rice, and logging each one separately is slow enough that people give up.

On top of that, getting enough protein without meat is a question vegetarians face every day, and generic apps don't help with it.

I designed and built Nourish on my own, from the UI to the database. You take a photo of your plate, and GPT-4o-mini's vision feature splits it into 4–8 separate food items to log. AI guesses are sometimes wrong, so one tap lets you correct an item, and each correction is saved as labelled data. On top of that sit personalised 7-day meal plans, AI meal swaps, protein targets, and streak and water tracking.

Behind the scenes, a response cache cut OpenAI costs by more than 60%. Per-user AI limits and Postgres row-level security keep each person's data private and stop anyone from running up the bill.

I built the full product before checking how often people would actually open it. Installs came quickly, but daily use didn't follow. Next time I'll put a rough version in front of ten real users before writing the backend. The engineering held up; the gap was in testing the idea first.