A personalised AI meal planner for plant-based eaters. Set your diet, goals, and preferences — the app generates a 7-day meal plan, tracks your pantry, and produces a shopping list for what's missing. Save any AI-generated meal you love with one click; future plans get smarter every time you do.

# Feature Description 1 Auth Sign in with Google via Firebase 2 Preferences Diet type, calorie target, excluded ingredients, free-text notes 3 Meal Plan Generator 7-day plan via Claude Sonnet, tabbed day view with nutrition strip 4 My Recipes Save any AI-generated meal with one click into your personal collection 5 Pantry + Shopping List Maintain pantry items; get a "what's missing" list on plan generation 6 Personalisation Behaviour signals feed a taste profile injected into every generation prompt

Layer Technology Frontend Next.js 15 (App Router), TypeScript strict, Tailwind CSS v4, ShadCN/UI, TanStack Query v5, Firebase Auth, React Hook Form + Zod Backend FastAPI, Pydantic v2, SQLAlchemy 2.0 async (asyncpg), Alembic AI — Generation Anthropic Claude Sonnet (vivid descriptions, strict JSON output) AI — Retrieval Gemini 2.0 Flash File Search (global + per-user recipe corpora) AI — Embeddings Gemini text-embedding-004 → pgvector (recipe semantic search only) Database PostgreSQL 16 + pgvector extension Cache Redis via Upstash (rate limiting) Auth Firebase Authentication (Google OAuth) Infra Google Cloud Run (API), Firebase Hosting (frontend), Cloud SQL, GCS, Secret Manager

Python 3.12+

Node.js 20+

Docker + Docker Compose

A Firebase project with Google OAuth enabled

A Firebase service account JSON (download from Firebase console → Project Settings → Service Accounts)

Edit backend/.env and fill in your values:

Place your Firebase service account file at backend/firebase-service-account.json (git-ignored).

Edit frontend/.env.local:

This starts PostgreSQL 16 with the pgvector extension available on localhost:5432.

This creates all 9 tables and enables the vector extension:

Table Purpose users Created on first Firebase login user_preferences Diet type, calorie target, exclusions meal_plans Full 7-day plan JSON blobs generated_meals Individual meals flattened from each plan (queryable rows) user_recipes Saved recipes — bookmarked AI meals (+ pgvector embedding column) pantry_items User's pantry shopping_lists Auto-generated "what's missing" lists user_signals Append-only behaviour event log (powers personalisation) user_taste_profiles Materialised taste profile rebuilt from signals; injected into every Claude prompt

Then open:

Frontend → http://localhost:3000

API docs → http://localhost:8000/docs

Health check → http://localhost:8000/health

All endpoints require Authorization: Bearer {firebase_id_token}.

Phase Status Phase 1 — Foundation (auth, health, DB, login page) ✅ Done Phase 2 — Data layer (all tables, migrations, preferences, pantry) 🔄 In progress Phase 3 — AI core (retrieval, generation, personalisation pipeline) ⬜ Pending Phase 4 — Meal plan UI (cards, tabs, bookmark flow) ⬜ Pending Phase 5 — Recipes, shopping, full signal coverage ⬜ Pending Phase 6 — Production deploy (Cloud Run, Cloud SQL, Firebase Hosting) ⬜ Pending