NutriAI

AI-powered nutrition intelligence — grounded in your own documents

Upload nutrition PDFs → Process & embed → Ask questions → Get precise, evidence-based answers

---

Overview

NutriAI is a production-ready full-stack SaaS platform that brings AI-powered question answering to your private nutrition knowledge base. Built on Retrieval-Augmented Generation (RAG), every answer is grounded in documents you upload — not model hallucinations.

┌─────────────┐    ┌──────────────┐    ┌────────────────┐    ┌──────────────┐
│  Upload PDF  │───▶│ Chunk + Embed│───▶│ pgvector Index │───▶│  LLM Answer  │
│  or TXT file │    │  (Cohere /   │    │  Vector Search │    │ (RAG-grounded│
│              │    │   OpenAI)    │    │  Similarity    │    │   response)  │
└─────────────┘    └──────────────┘    └────────────────┘    └──────────────┘

---

Features

• RAG Pipeline — Upload → Chunk → Embed → Index → Retrieve → Generate
• Semantic Search — Vector similarity search via pgvector or Qdrant
• Multi-LLM Support — Cohere (command-a-03-2025) or OpenAI, switchable via env
• JWT Authentication — Secure register/login with 7-day tokens
• Premium UI — Dark-mode SaaS interface built with Framer Motion animations
• Single-port Production — FastAPI serves the compiled React build on one port
• Private Knowledge Base — Per-user, per-project document isolation
• Real-time Processing Feedback — Step-by-step upload → chunk → index progress UI
• Swagger & ReDoc — Auto-generated API documentation at /api/docs
• Metrics — Prometheus metrics via starlette-exporter

---

Tech Stack

Backend

Layer	Technology
Framework	FastAPI 0.110 + Uvicorn
ORM	SQLAlchemy 2.0 (async) + asyncpg
Vector DB	pgvector (PostgreSQL) · Qdrant
LLM / Embeddings	Cohere 5.6 · OpenAI 2.2
RAG	LangChain 0.1
Auth	python-jose (JWT) · bcrypt 4.1
PDF Parsing	PyMuPDF 1.24
Monitoring	Prometheus · starlette-exporter

Frontend

Layer	Technology
Framework	React 18 + Vite 5
Routing	React Router 6
Styling	Tailwind CSS 3
Animations	Framer Motion 11
Icons	Lucide React
HTTP	Axios
Markdown	react-markdown
File Upload	react-dropzone
Notifications	react-hot-toast

---

Project Structure

nutriai/
├── src/                          # FastAPI backend
│   ├── main.py                   # App factory, CORS, routers, static serving
│   ├── .env.example              # Environment variable template
│   ├── requirements.txt
│   ├── routes/
│   │   ├── auth.py               # POST /register, /login  GET|PUT /me
│   │   ├── data.py               # POST /upload, /process
│   │   ├── nlp.py                # POST /index/push, /answer  GET /index/info
│   │   └── schema/               # Pydantic request/response models
│   ├── controllers/
│   │   ├── DataController.py     # File validation & path management
│   │   ├── NLPController.py      # RAG orchestration
│   │   └── ProcessController.py  # Chunking logic (do not modify)
│   ├── models/
│   │   ├── db_schemas/           # SQLAlchemy ORM models
│   │   │   └── user.py           # User model (auto-created on startup)
│   │   └── ProjectModel.py       # Project creation and retrieval
│   └── helpers/
│       └── config.py             # Pydantic Settings (env vars)
│
├── frontend/                     # React + Vite SPA
│   ├── src/
│   │   ├── App.jsx               # Router + auth guards
│   │   ├── index.css             # Tailwind + design system tokens
│   │   ├── context/
│   │   │   └── AuthContext.jsx   # JWT state management
│   │   ├── services/
│   │   │   └── api.js            # Axios client + interceptors
│   │   ├── components/
│   │   │   └── Layout.jsx        # Sidebar shell + route transitions
│   │   └── pages/
│   │       ├── Landing.jsx       # Public marketing page
│   │       ├── Login.jsx         # Split-screen auth
│   │       ├── Register.jsx      # Split-screen auth
│   │       ├── Dashboard.jsx     # Stats + quick actions
│   │       ├── Chat.jsx          # RAG chat interface
│   │       ├── Documents.jsx     # Upload + process + index
│   │       └── Profile.jsx       # Account management
│   └── dist/                     # Production build (served by FastAPI)
│
├── start.sh                      # Production startup script
├── dev.sh                        # Development startup (two servers)
├── .gitignore
└── README.MD

---

Quick Start

Prerequisites

• Python 3.12+
• Node.js 18+
• PostgreSQL with pgvector extension
• Cohere API key or OpenAI API key

1 — Clone & install

git clone https://github.com/Youssefx64/End-to-End-Nutrition-RAG-System.git
cd End-to-End-Nutrition-RAG-System

# Python dependencies
pip install -r src/requirements.txt

# Node dependencies
cd frontend && npm install && cd ..

2 — Configure environment

cp src/.env.example src/.env

Open src/.env and fill in:

# Required
POSTGRES_USERNAME=postgres
POSTGRES_PASSWORD=your_password
POSTGRES_HOST=localhost
POSTGRES_PORT=5432
POSTGRES_MAIN_DATABASE=nutriai

# Generate a strong secret:  python3 -c "import secrets; print(secrets.token_hex(32))"
JWT_SECRET_KEY=your_random_secret_here

# Pick one LLM provider
COHERE_API_KEY=your_cohere_key_here
# OPENAI_API_KEY=your_openai_key_here
# GENERATION_BACKEND=OPENAI

3 — Run

Production mode (single port 5000 — builds React then serves via FastAPI):

bash start.sh
# → http://localhost:5000

Development mode (hot-reload on both frontend and backend):

bash dev.sh
# Frontend → http://localhost:5000  (Vite HMR)
# Backend  → http://localhost:8000  (uvicorn --reload)

---

API Reference

Base URL: http://localhost:5000/api/v1

Interactive docs: /api/docs (Swagger UI) · /api/redoc

Authentication

Method	Endpoint	Description	Auth
POST	/auth/register	Create account → returns JWT	—
POST	/auth/login	Sign in → returns JWT	—
GET	/auth/me	Get current user	Bearer
PUT	/auth/me	Update profile	Bearer

Data Pipeline

Method	Endpoint	Description	Auth
POST	/data/upload/{project_id}	Upload PDF or TXT file	Bearer
POST	/data/process/{project_id}	Chunk file into DataChunks	Bearer

Process request body:

{
  "file_id": "abc123.pdf",
  "chunk_size": 512,
  "overlap_size": 50,
  "do_reset": 0
}

NLP / RAG

Method	Endpoint	Description	Auth
POST	/nlp/index/push/{project_id}	Embed chunks → push to vector DB	Bearer
GET	/nlp/index/info/{project_id}	Vector DB collection stats	Bearer
POST	/nlp/index/search/{project_id}	Semantic similarity search	Bearer
POST	/nlp/index/answer/{project_id}	RAG question answering	Bearer

Answer request body:

{
  "text": "What are the best plant-based protein sources?",
  "limit": 5
}

Answer response:

{
  "signal": "rag_answer_success",
  "answer": "Based on your documents, the top complete plant proteins are...",
  "full_prompt": "...",
  "chat_history": []
}

---

RAG Architecture

User Question
      │
      ▼
┌─────────────────────────────────────────────────────┐
│                   NLPController                      │
│                                                     │
│  1. Embed question  ──▶  Cohere/OpenAI Embeddings   │
│                                                     │
│  2. Vector search   ──▶  pgvector / Qdrant          │
│                          (cosine similarity, top-K) │
│                                                     │
│  3. Build prompt    ──▶  Retrieved chunks           │
│                          + question template        │
│                                                     │
│  4. Generate        ──▶  Cohere command-a-03-2025   │
│                          or OpenAI GPT              │
└─────────────────────────────────────────────────────┘
      │
      ▼
  Grounded Answer

---

Environment Variables

Variable	Required	Default	Description
JWT_SECRET_KEY	✅	—	Random secret for JWT signing
ALLOWED_ORIGINS	—	*	CORS origins (comma-separated)
POSTGRES_USERNAME	✅	—	PostgreSQL username
POSTGRES_PASSWORD	✅	—	PostgreSQL password
POSTGRES_HOST	✅	localhost	PostgreSQL host
POSTGRES_PORT	✅	5432	PostgreSQL port
POSTGRES_MAIN_DATABASE	✅	—	Database name
GENERATION_BACKEND	—	COHERE	COHERE or OPENAI
EMBEDDING_BACKEND	—	COHERE	COHERE or OPENAI
COHERE_API_KEY	✅*	—	Cohere API key
OPENAI_API_KEY	✅*	—	OpenAI API key
GENERATION_MODEL_ID	—	command-a-03-2025	LLM model identifier
EMBEDDING_MODEL_ID	—	embed-multilingual-v3.0	Embedding model
VECTOR_DB_BACKEND	—	PGVECTOR	PGVECTOR or QDRANT
FILE_MAX_SIZE	—	10	Max upload size in MB

*One of COHERE_API_KEY or OPENAI_API_KEY is required.

---

Pages

Route	Description
/	Landing page — features, how it works, CTA
/register	Create account
/login	Sign in
/dashboard	Knowledge base stats, quick actions, getting started
/chat	AI chat — RAG-powered Q&A against your documents
/documents	Upload, process, and index documents
/profile	Edit profile, security info

---

Production Checklist

Before deploying:

• Set JWT_SECRET_KEY to a cryptographically strong random value
• Set ALLOWED_ORIGINS to your actual domain (not *)
• Set your COHERE_API_KEY or OPENAI_API_KEY
• Ensure src/.env is not committed (protected by .gitignore)
• Configure PostgreSQL with the pgvector extension enabled

Generate a strong secret key:

python3 -c "import secrets; print(secrets.token_hex(32))"

Enable pgvector in PostgreSQL:

CREATE EXTENSION IF NOT EXISTS vector;

---

Contributing

Contributions are welcome! Here's how to get started:

# 1. Fork the repository
# 2. Create a feature branch
git checkout -b feature/your-feature-name

# 3. Make your changes, following existing code patterns
# 4. Test your changes
bash dev.sh

# 5. Commit with a clear message
git commit -m "feat: add your feature description"

# 6. Push and open a Pull Request
git push origin feature/your-feature-name

Important: Do not modify src/controllers/ProcessController.py — this file is part of the core RAG pipeline and is intentionally preserved as-is.

---

License

MIT License — see LICENSE for details.

---

Built with FastAPI · React · pgvector · Cohere

Live Demo · Report Bug · Request Feature