🤖 Satya RAG — AI-Powered Document Assistant

A full-stack Retrieval-Augmented Generation (RAG) application that allows users to upload documents and ask questions about them using AI. The system combines semantic search with a large language model to provide accurate, context-aware responses streamed in real-time.

---

📸 Architecture Overview

┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│   Frontend   │────▶│  Backend API │────▶│    Redis     │
│   (Next.js)  │     │  (Laravel)   │     │  (Cache +    │
│   Port 3000  │     │   Port 8000  │     │   History)   │
└──────┬───────┘     └──────────────┘     └──────────────┘
       │
       │  Streaming Chat
       ▼
┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│  AI Backend  │────▶│   Pinecone   │     │    Ollama    │
│  (FastAPI)   │     │ (Vector DB)  │     │ (Embeddings) │
│   Port 8081  │     └──────────────┘     │  Port 11434  │
└──────┬───────┘                          └──────────────┘
       │
       ▼
┌──────────────┐
│  Google AI   │
│  (Gemini)    │
└──────────────┘

---

✨ Features

💬 AI Chat

• Streaming responses — real-time token-by-token output like ChatGPT
• Markdown rendering — bold, tables, code blocks, lists, headings rendered in chat
• Semantic cache — identical/similar questions are answered instantly from Redis cache
• Conversation history — stored per session in Redis with 24h TTL
• Re-ranking — retrieves 10 documents, re-ranks to top 4 using FlashRank for accuracy
• Session management — create, rename, and delete chat conversations

📄 Document Management

• Upload documents — supports PDF, DOCX, TXT, and Markdown files
• Automatic chunking — documents are split with RecursiveCharacterTextSplitter
• Vector embeddings — chunks are embedded using Ollama (mxbai-embed-large) and stored in Pinecone
• Delete documents — remove documents and their vectors from Pinecone

📊 Analytics Dashboard

• Total tokens used, total chats, active users
• Cache hit rate monitoring
• Hourly token usage trend (last 24 hours)
• Top FAQ topics
• Recent chat logs

🔐 Authentication

• Email/password login via Laravel Sanctum
• Google OAuth login via Laravel Socialite
• Protected routes with middleware

🎨 UI/UX

• Dark/light mode toggle
• Responsive sidebar with session management
• Smooth streaming with RAF-based rendering (no jitter)
• Loading states and micro-animations

---

🛠️ Tech Stack

Layer	Technology
Frontend	Next.js 16, React 19, TypeScript, Tailwind CSS v4, Zustand, Framer Motion
Backend API	Laravel 12, PHP 8.2, Sanctum, Socialite
AI Backend	FastAPI, LangChain, Python 3.10
LLM	Google Gemini (via langchain-google-genai)
Embeddings	Ollama (mxbai-embed-large)
Vector DB	Pinecone
Cache & History	Redis (Redis Stack with vector search)
Re-ranking	FlashRank
Containerization	Docker & Docker Compose

---

📋 Prerequisites

Before you begin, make sure you have the following installed:

• Node.js ≥ 20.x
• PHP ≥ 8.2 + Composer
• Python ≥ 3.10
• Docker & Docker Compose
• Ollama running locally with mxbai-embed-large model pulled
• Pinecone account
• Google AI API Key (for Gemini)

---

🚀 Setup & Installation

1. Clone the repository

git clone https://github.com/SatyaFebi/NEW_RAG.git
cd NEW_RAG

2. Setup Ollama (Embeddings)

# Install Ollama: https://ollama.com/download
ollama pull mxbai-embed-large
ollama serve  # Runs on port 11434

3. Setup AI Backend (FastAPI)

# Copy environment file
cp backend-fastapi/.env.example backend-fastapi/.env

Edit backend-fastapi/.env with your credentials:

PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX_NAME=your_index_name
EMBEDDING_MODEL_NAME=mxbai-embed-large
GOOGLE_API_KEY=your_google_ai_api_key
GEMINI_MODEL=gemini-2.5-flash-lite
REDIS_URL=redis://redis:6379

# Optional: LangSmith tracing
LANGSMITH_TRACING=false
LANGSMITH_API_KEY=
LANGSMITH_PROJECT=

4. Start Docker Services (FastAPI + Redis)

docker compose up -d --build

This starts:

• FastAPI AI backend on http://localhost:8081
• Redis Stack on http://localhost:6379

5. Setup Laravel Backend

cd backend
cp .env.example .env
composer install
php artisan key:generate
php artisan migrate
php artisan serve  # Runs on port 8000

6. Setup Frontend (Next.js)

cd frontend
npm install

Create/edit frontend/.env:

NEXT_PUBLIC_API_URL=http://localhost:8000/api
NEXT_PUBLIC_AI_API_URL=http://localhost:8081

npm run dev  # Runs on port 3000

7. Open the App

Navigate to http://localhost:3000 — login and start chatting!

---

🏗️ Production Build

Frontend

cd frontend
npm run build
npm start  # Serves production build on port 3000

Docker (Full Stack)

Uncomment the laravel_app and nextjs_web services in docker-compose.yml, then:

docker compose up -d --build

---

📁 Project Structure

NEW_RAG/
├── backend/                 # Laravel API (Auth, User Management)
│   ├── app/
│   ├── routes/api.php       # API routes (login, user, OAuth)
│   ├── .env.example
│   └── ...
│
├── backend-fastapi/         # FastAPI AI Service
│   ├── main.py              # All AI logic (chat, upload, dashboard)
│   ├── requirements.txt     # Python dependencies
│   ├── Dockerfile
│   └── .env.example
│
├── frontend/                # Next.js Frontend
│   ├── src/
│   │   ├── app/
│   │   │   ├── chat/        # Chat page with streaming
│   │   │   ├── dashboard/   # Analytics dashboard
│   │   │   ├── documents/   # Document management
│   │   │   ├── login/       # Login page
│   │   │   └── globals.css  # Design system + markdown styles
│   │   ├── components/      # Sidebar, ThemeProvider
│   │   ├── store/           # Zustand stores (chat, auth, theme)
│   │   └── lib/             # API utilities
│   └── .env
│
├── docker-compose.yml       # Orchestrates FastAPI + Redis
└── README.md

---

🔧 API Endpoints (FastAPI)

Method	Endpoint	Description
GET	/	Health check
POST	/chat	Send message, receive streaming AI response
POST	/documents/upload	Upload document (PDF, DOCX, TXT, MD)
GET	/documents	List documents (limited)
DELETE	/documents/{doc_id}	Delete document by ID
GET	/dashboard/stats	Get analytics data

---

🔧 API Endpoints (Laravel)

Method	Endpoint	Description
POST	/api/login	Email/password login
GET	/api/user	Get authenticated user (Sanctum)
GET	/api/auth/google	Redirect to Google OAuth
GET	/api/auth/google/callback	Handle Google OAuth callback

---

🐛 Troubleshooting

Problem	Solution
Ollama connection refused	Make sure ollama serve is running on port 11434
Pinecone timeout	Check your API key and index name in .env
Redis connection error	Ensure Redis container is running: docker ps
Frontend not updating	Run npm run dev or npm run build
FastAPI container error	Check logs: docker logs fastapi_rag
Vite manifest error	Run npm run build in /frontend

---

📝 License

This project is for educational and personal use.

---

Built with ❤️ by Satya