| # Vietnamese Legal Chatbot - Presentation Setup Guide | |
| ## Quick Start for Presentation | |
| ### Prerequisites | |
| 1. **Python 3.8+** installed | |
| 2. **Docker** installed and running | |
| 3. **Google API Key** for Gemini (optional for demo) | |
| ### Step 1: Install Dependencies | |
| ```bash | |
| pip install -r requirements.txt | |
| ``` | |
| ### Step 2: Start Qdrant Database | |
| ```bash | |
| python start_qdrant.py | |
| ``` | |
| This will: | |
| - Pull Qdrant Docker image | |
| - Start Qdrant on http://localhost:6333 | |
| - Create persistent storage in `qdrant_data/` folder | |
| ### Step 3: Set Up the System | |
| ```bash | |
| python setup_system.py | |
| ``` | |
| This will: | |
| - Load 3,271 legal documents | |
| - Create 61,068 document chunks | |
| - Build vector and BM25 indices | |
| - Set up the RAG system | |
| ### Step 4: Run the Application | |
| ```bash | |
| python app.py | |
| ``` | |
| This will: | |
| - Start the Gradio web interface | |
| - Open at http://localhost:7860 | |
| - Show initialization progress | |
| ## Demo Questions for Presentation | |
| ### Sample Legal Questions to Try: | |
| 1. **"Điều kiện thành lập doanh nghiệp là gì?"** | |
| - Tests basic legal knowledge retrieval | |
| 2. **"Quy định về thời gian làm việc tối đa trong ngày?"** | |
| - Tests labor law knowledge | |
| 3. **"Thủ tục đăng ký kết hôn cần những gì?"** | |
| - Tests civil law procedures | |
| 4. **"Mức phạt vi phạm giao thông đường bộ?"** | |
| - Tests administrative law | |
| ## Presentation Structure (10-12 minutes) | |
| ### 1. Introduction (2 min) | |
| - **Problem**: Legal information access in Vietnam | |
| - **Solution**: AI-powered legal assistant using RAG | |
| - **Technology**: Hybrid search (BM25 + Vector) + LLM | |
| ### 2. Technical Architecture (3 min) | |
| - Show the system components | |
| - Explain hybrid retrieval approach | |
| - Highlight Vietnamese-specific optimizations | |
| ### 3. Live Demo (3 min) | |
| - Show the web interface | |
| - Ask sample questions | |
| - Demonstrate response quality and citations | |
| ### 4. Performance Results (2 min) | |
| - Show performance table from `results_table.txt` | |
| - Highlight 60.82% MRR achievement | |
| - Compare different methods | |
| ### 5. Future Work (1 min) | |
| - Expand legal corpus | |
| - Mobile app development | |
| - Integration with legal services | |
| ## Troubleshooting | |
| ### If Qdrant fails to start: | |
| ```bash | |
| # Check Docker status | |
| docker ps | |
| # Restart Qdrant | |
| python start_qdrant.py stop | |
| python start_qdrant.py | |
| ``` | |
| ### If setup fails: | |
| ```bash | |
| # Clean up and retry | |
| rm -rf qdrant_data/ | |
| python start_qdrant.py | |
| python setup_system.py | |
| ``` | |
| ### If app fails to start: | |
| - Check if Google API key is set (optional) | |
| - Ensure Qdrant is running on port 6333 | |
| - Check console for error messages | |
| ## Key Features to Highlight | |
| 1. **Hybrid Search**: Combines keyword (BM25) and semantic (vector) search | |
| 2. **Vietnamese-Specific**: Uses specialized Vietnamese embedding models | |
| 3. **Reranking**: Advanced document re-ranking for better relevance | |
| 4. **Real-time Interface**: Gradio web interface with progress indicators | |
| 5. **Source Attribution**: Always cites specific legal documents | |
| 6. **Fallback System**: Can search Google if local documents insufficient | |
| ## Performance Metrics | |
| - **Best Method**: Hybrid 2 + Reranking | |
| - **MRR**: 60.82% | |
| - **Coverage**: 88.99% | |
| - **Response Time**: ~0.6 seconds | |
| - **Documents**: 3,271 legal documents, 61,068 chunks |