Neural nutrition recommender

AI Nutritionist

Public RepoML RecommenderFastAPI95 Tests

AI Nutritionist is a local-first nutrition planner that generates daily and weekly meal plans from a processed USDA/FNDDS catalog plus curated Mediterranean/Greek rows and a small reviewed recipe-backed pilot. I rebuilt it as a public software project with profile-aware targets, bounded weight-loss calorie targets, Hybrid Recommender V2 ranking, weekly planner summaries, diet-mode guardrails, local session feedback, flat grocery export, ingredient-level recipe grocery export for pilot rows, Streamlit/CLI/API surfaces, model/data cards, verified Docker runtime health, CI/security readiness, and 95 pytest tests. The output is positioned as general wellness guidance, not medical advice.

What it proves

Generates daily and weekly meal plans with profile targets, weekly planner summaries, bounded weight-loss targets, diet-mode guardrails, alternatives, flat grocery export, recipe ingredient export, and local feedback.

What it proves

Shows data processing, recipe validation/projection, Hybrid Recommender V2 ranking, Streamlit/CLI/API delivery, model/data cards, Docker runtime health proof, CI/security readiness, GitHub Actions CI, and 95 pytest tests in a reviewer-friendly local app.

GitHub Download ZIP

Impact

Generates daily and weekly meal plans with profile targets, weekly planner summaries, bounded weight-loss targets, diet-mode guardrails, alternatives, flat grocery export, recipe ingredient export, and local feedback.
Shows data processing, recipe validation/projection, Hybrid Recommender V2 ranking, Streamlit/CLI/API delivery, model/data cards, Docker runtime health proof, CI/security readiness, GitHub Actions CI, and 95 pytest tests in a reviewer-friendly local app.
Keeps wellness output safely scoped without medical claims.

Problem

Nutrition demos often stop at a generic recommender. This project needs to show data processing, profile-aware constraints, transparent recommendations, and safety limits in one reviewable local app.

Approach

The app combines a processed USDA/FNDDS catalog, curated Mediterranean/Greek foods, deterministic target estimation, Hybrid Recommender V2 ranking, hard meal-planning guardrails, and local UI/CLI surfaces.

Current status

Public project with generated-plan screenshots, weekly planner summaries, recipe-backed pilot data, expanded tests, local demo surfaces, verified Docker runtime health, CI/security readiness, API payloads, model/data cards, and clear wellness boundaries.

Architecture / workflow

Data-processing layer prepares a reusable food catalog and curated extension.
Recipe layer validates source/recipe/ingredient CSVs, aggregates nutrients deterministically, and projects reviewed recipe rows into the public catalog schema.
Profile layer estimates calorie and macro targets with bounded controls.
Ranking layer scores meals with Hybrid Recommender V2, bounded substitutions, portion adjustments, and public payload boundaries.
Planning layer applies meal slots, diet modes, exclusions, and safety guardrails.
Streamlit, CLI, and FastAPI surfaces expose the same local recommendation behavior while public payload serializers hide internal scores and keep feedback local-first.

Next steps

Keep screenshots current after UI improvements.
Add a concise architecture diagram if the page needs more visual explanation.
Use this level of evidence as the bar for the rest of the portfolio.

AI Nutritionist generated daily meal plan with meal title, nutrient metrics, and food table

AI Nutritionist generated weekly Mediterranean-style meal rotation

AI Nutritionist generated daily nutrition progress view

AI Nutritionist generated swap alternatives for meal components

AI Nutritionist mobile generated day-detail view

AI Nutritionist Streamlit setup screen before generation