Research areas
AI for nutrition & healthy living
Helping people eat better and move more, safely: understanding Vietnamese food descriptions, calorie figures from sourced data, and safety checks before any advice.
Most nutrition apps know Western food far better than phở, cơm tấm or trà sữa, and many give advice without asking whether it is safe for the person.
What we work on:
• Understanding how Vietnamese people describe meals: portion words (bát, tô, đĩa, ly), sizes, quantities written as words, and the accents that tell "muống" (water spinach) from "muỗng" (spoon).
• A food table built from USDA FoodData Central (public domain) and standard Vietnamese recipes, where every calorie and nutrient figure is computed from sourced ingredients and reviewed by a partner nutrition expert.
• Safety first: a screening step before any weight-loss plan, minimum calorie levels and maximum weekly loss rates it never exceeds, referrals to a doctor for people for whom a plan isn't safe, and emergency replies that don't depend on a model.
• Privacy for health data: explicit consent, export and deletion on request, and no use for model training.
This work powers eChat's health coach, now in development.
Used in eChat