Neural nutrition recommender
AI Nutritionist
AI Nutritionist is a local-first nutrition recommendation system, presented as a standalone public software project rather than a thesis or academic submission. It uses a processed USDA FoodData Central / FNDDS catalog, a curated Mediterranean/Greek extension, profile-aware target estimation, explicit maintain/lose/gain weight-goal controls, optional lean-mass-aware protein targets, a locally trained neural MLP food ranker, weekly Mediterranean-style rotation across poultry, fish, legumes, vegetables, whole grains/starches and olive-oil sides, conservative vegan/vegetarian filtering, keto-style low-carb guardrails, preference-aware avoid/prefer controls, nutrition-focus modes, meal group diversity, Plan Fit scoring, alternatives, macro split reporting, and explanation text while keeping clear boundaries that the output is general wellness guidance, not medical advice.
Highlights
- Builds daily and weekly meal plans from protein, produce, whole-grain/starch, and healthy-fat slots.
- Trains a deterministic neural ranker from USDA-derived weak labels, then applies hard meal guardrails and user preference controls.
- Adds practical Mediterranean/Greek meals plus vegan, vegetarian and keto-style modes without overclaiming clinical or medical accuracy.
Validation
40 pytest tests
BMI/age/diet evaluation matrix
CLI Mediterranean weekly smoke test
Streamlit local smoke
Research notes
