Forma
Product Designer & AI Systems Engineer · Independent
Problem: Western fitness apps are built for Western diets — they fail Indian users eating dal, roti, and home-cooked sabzi, and impose rigid plans that trigger guilt and churn when real life gets in the way.
What I built:
- →Designed an LLM orchestration pipeline (Google Gemini API) that analyzes a 7-day rolling window of real user adherence to dynamically recalculate daily calories, macros, and workout intensity.
- →Built Indian diet-aware nutrition logic — recognizing regional foods and portion patterns instead of defaulting to Western food databases.
- →Implemented deterministic fallback rules with structured JSON schema outputs to ensure consistent, safe coaching responses even on LLM edge cases.
- →Delivered a full-stack adaptive coaching experience — adapting plans to off-days instead of penalizing them, reducing user guilt and improving retention.
Result: A live AI fitness coach purpose-built for Indian diets and routines — with adaptive daily guidance that evolves with the user's real behaviour, not an ideal one.