NutriSight

NutriSight is a local-first FastAPI web app for photo-based nutrition logging with a mobile-first multi-page UI. It supports:

• scan a meal photo and get AI-detected food items with editable macros
• manually log foods via search against a 7,800+ item nutrition database
• AI food lookup: get nutrition estimates for any food via LLM, with optional web search
• edit AI estimates (name, macros) and save to the database for future use
• save favorite foods for one-tap re-logging
• review and edit meals after saving (update items, macros, delete)
• daily dashboard with calorie and macro tracking
• AI-powered meal recommendations for the rest of the day (Indian vegetarian default)
• full meal history with trends, grouped by day
• per-user accounts with persistent device sessions
• configurable AI provider (LM Studio, Ollama, API, or stub)

Pages

The app is organized into focused mobile-first pages with a bottom tab bar:

Page	Path	Purpose
Dashboard	/	Today's calories/macros, AI meal suggestions, recent meals with delete
Scan	/analyze	Photo capture, AI analysis, review items with editable macros
Quick Log	/log	Search foods, AI food lookup, build meals manually, manage favorites
History	/history	Day-grouped meal history, edit/delete saved meals
Settings	/settings	Goals, AI provider config, account management
Admin DB	/admin/db	Nutrition catalog search/create/update/delete
Admin Users	/admin/users	User management

Stack

• FastAPI + Jinja2 server-rendered templates
• Vanilla JavaScript with per-page modules (shared.js, analyze.js, log.js)
• SQLite for meals, nutrition catalog, users, and settings
• LM Studio integration (OpenAI-compatible API) for vision detection and portion estimation
• Direct LM Studio calls from browser for meal recommendations (proxied through /api/llm/chat to avoid CORS)

Run

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Quick start (kills any existing server on port 8000 and restarts)
./host.sh

# Or manually
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Then open http://localhost:8000 from your laptop or http://<your-lan-ip>:8000 from your phone on the same network.

Test

pytest tests/ -v

57 tests covering:

• Nutrition math, daily aggregation, and user-scoped trend helpers
• API endpoint validation (save meals, search, admin CRUD, LLM proxy)
• AI food lookup (input validation, LLM estimation, DB persistence)
• LM Studio payload parsing and service integration
• Playwright UI tests for the AI Help modal (visibility, interaction, edit flow, save-to-DB integration)

Key Features

Meal Analysis (Scan Page)

• Camera capture with client-side image compression
• AI detection via LM Studio with Qwen3-VL
• Item cards show all macros (P/C/F) and calories
• Tap macros to edit via popup; changes override auto-calculated values
• S/M/L portion toggle + grams stepper
• Change food mapping via full-text search
• Undo on item removal

AI Meal Recommendations (Dashboard)

• Calls LM Studio directly to suggest meals for remaining meal slots
• Accounts for remaining calories, protein, carbs, and fat
• Time-aware: breakfast (8am), lunch (12pm), snack (4pm), dinner (7pm)
• 2 options per meal slot with ingredient weights
• Dynamic per-meal calorie allocation that sums to the full budget
• Adjustable calorie budget with −/+ buttons (100 kcal steps)
• Indian vegetarian by default; override with keyword input
• Refresh button to regenerate suggestions
• Hidden after 9pm (no meals suggested)

AI Food Lookup (Quick Log Page)

• Floating "AI Help" button opens a centered modal
• Type any food or dish name to get AI-estimated nutrition (calories, protein, carbs, fat)
• Optional web search toggle queries DuckDuckGo for a second data point
• Edit button lets you customize the food name and all macro values before saving
• "Use in meal" adds the item (original or edited) to the meal builder
• "Save to DB" persists to the nutrition database so it's searchable next time
• Refresh button re-runs the lookup
• Input validated against injection attacks (regex whitelist, max 200 chars)

Meal History (History Page)

• 14-day bar chart and day-grouped meal list
• Tap to expand meal detail with per-item macros
• Edit button: inline editing of meal name, grams, calories, and all macros per item
• Delete items or entire meals
• Top logged foods section

Dashboard

• Dark card with today's calorie/macro summary and remaining budgets (protein, carbs, fat)
• Hover/touch on recommendation cards previews how remaining budgets would change
• Quick action buttons (Scan, Quick Log)
• Recent meals with inline delete
• AI meal plan section with adjustable calorie budget

API Endpoints

User-Facing

• GET / — Dashboard
• GET /analyze — Scan meal page
• GET /log — Quick log page
• GET /history — Meal history page
• GET /settings — Settings page
• POST /api/analyze — Upload image and run analysis
• POST /api/meals — Save a meal
• GET /api/meals/{id} — Fetch meal detail
• PUT /api/meals/{id} — Update a saved meal (name, items, macros)
• DELETE /api/meals/{id} — Delete a meal
• GET /api/foods — Search nutrition catalog
• POST /api/settings — Update goals and provider config
• POST /api/llm/chat — Proxy chat requests to LM Studio
• POST /api/ai-food-lookup — AI-powered nutrition estimation for any food (with optional web search)
• POST /api/ai-food-lookup/save — Save an AI-looked-up food to the nutrition database
• POST /custom-foods — Create a user-owned custom food
• POST /custom-foods/{id}/log — Log a custom food as a meal
• POST /custom-foods/{id}/delete — Delete a custom food

Admin

• GET /admin/db — Nutrition catalog management
• GET /admin/users — User management
• POST /admin/nutrition-items — Create/update nutrition item
• POST /admin/nutrition-items/{id}/delete — Delete nutrition item
• POST /admin/label-import — Import from nutrition label image

GitHub Pages Preview

Static rendered previews of all pages with sample data:

• https://apoorva87.github.io/NutriVisionAI/

What the preview is:

• static rendered preview of the real templates with sample data
• useful for checking layout and mobile presentation

What the preview is not:

• not the live FastAPI app; forms, uploads, and API calls do not work

How to enable:

1. Go to repo Settings → Pages
2. Set Source to GitHub Actions
3. Push to main or manually run the Deploy Template Preview workflow

Files:

• Workflow: .github/workflows/pages.yml
• Renderer: scripts/render_template_previews.py
• Output: site/

LM Studio Setup

1. Start LM Studio on your machine (default: http://localhost:1234)
2. Load a vision-language model (e.g., qwen/qwen3-vl-8b)
3. In NutriSight Settings, choose lmstudio as provider and configure the URL and model
4. Save settings, then scan a meal

Project Docs

• ARCHITECTURE.md — Runtime flow, route structure, DB schema, code layout
• SIGNON.md — Auth options and deployment guidance
• TODO.md — Current backlog
• docs/EXECUTION_LOG.md — Session-by-session handoff log
• docs/ROADMAP.md — Implementation sequence

Nutrition Catalog

The local catalog stores 7,849 items from USDA Foundation, USDA SR Legacy, and Indian (IFCT) sources. Bootstrap from scratch:

.venv/bin/python scripts/bootstrap_nutrition_sources.py --download --extract --verify
.venv/bin/python scripts/import_local_catalog.py