RAG Chat Application

A full-stack Retrieval-Augmented Generation (RAG) chat application that combines LLM capabilities with document-based knowledge retrieval. Upload documents, build a knowledge base, and chat with AI using context-aware responses.

Visit site: https://zhirong-chat.zeabur.app/

Features

Core Features

• RAG-Enhanced Chat: Chat with AI using standard LLM responses or RAG-enhanced responses with document context
• LangGraph Agent System: Multi-step reasoning with integrated search capabilities (Google via Serper, Tavily Search)
• Document Management: Upload and manage PDF, TXT, Markdown, DOCX, JSON, CSV, and more
• Multi-Provider LLM Support: Gemini, OpenAI, and Anthropic API key options
• Gemini File Search: Automatic document indexing, embedding, and semantic retrieval with per-user stores

Authentication & Security

• User Authentication: JWT-based authentication with bcrypt password hashing
• Email Verification: Required email verification for new user accounts via Gmail SMTP
• Password Reset: Secure password reset flow with email-based JWT tokens

Development & Deployment

• Real-time Updates: Hot reload for development (frontend and backend)
• Database Migrations: Alembic-based migration system for schema changes
• Form Validation: VeeValidate with Yup schemas for robust input validation
• Comprehensive Testing: pytest (backend), Vitest (frontend unit), Playwright (E2E)
• Dockerized: Easy deployment with Docker Compose (dev and production modes)
• Utility Scripts: Makefile and shell scripts for common operations
• Automated Deployment: Continuous deployment via Zeabur - automatically builds and deploys on every push to GitHub

Screenshots

LangGraph Workflow

The agent uses a multi-node graph for intelligent processing:

┌──────────┐
│ Analyzer │ - Analyzes query and decides if search is needed
└────┬─────┘
     │
     ├─── needs_search? ───┐
     │                     │
     ▼                     ▼
┌────────┐           ┌─────────┐
│ Search │           │ Respond │
└───┬────┘           └─────────┘
    │
    ▼
┌──────────┐
│ Responder│ - Generates final response with search context
└──────────┘

LLM Response with Web Search Results

RAG Implementation with Gemini File Search API

Tech Stack

Backend

• Framework: FastAPI with Uvicorn/Gunicorn
• Database: PostgreSQL 16 with SQLModel ORM
• Caching & Rate Limiting: Redis 7 with sliding window rate limiter
• Authentication: JWT tokens with bcrypt password hashing
• AI/ML:
  • LLM Providers: Gemini, OpenAI, Anthropic (via Langchain)
  • Agent Framework: LangGraph for multi-step reasoning
  • RAG: Gemini File Search API
  • Search Tools: Serper (Google Search), Tavily Search
• Email: Gmail SMTP for verification and password reset
• Migrations: Alembic for database schema management
• Testing: pytest with coverage

Frontend

• Framework: Vue 3.5 with TypeScript 5.9
• Build Tool: Vite 7.1
• State Management: Pinia 3.0
• Routing: Vue Router 4.5
• UI Framework: Bootstrap 5.3 with Bootstrap Icons
• Form Validation: VeeValidate 4.15 with Yup
• Markdown: Marked 17.0 with Highlight.js 11.11 for code highlighting
• Testing: Vitest (unit), Playwright (E2E)

Infrastructure

• Containerization: Docker with Docker Compose
• Caching: Redis 7 (Alpine) for rate limiting and session management
• Web Server: Nginx (production)
• Deployment: Zeabur - Serverless cloud platform with GitHub integration for automated CI/CD

Utility Scripts

The scripts/ directory contains helpful tools for development and operations:

Redis Monitoring

Redis is used for rate limiting (20 messages per 30 minutes per user).

Quick Monitor (Shell Script):

./scripts/redis-monitor.sh

Other Scripts

• scripts/dev.sh - Start development environment
• scripts/prod.sh - Start production environment
• scripts/logs.sh - View application logs
• scripts/health-check.sh - Check service health
• scripts/db-backup.sh / scripts/db-restore.sh - Database backup/restore