Docbot - AI-Powered Document Assistant 🤖

Overview 🔍

Docbot is an intelligent document assistant that processes and analyzes documents to provide accurate answers to user queries. It leverages multiple advanced language models and retrieval techniques to ensure high-quality responses.

Features ⭐

• Multi-document processing and analysis
• Multi-route retrieval system for improved accuracy
• Advanced reranking mechanism
• Interactive chat interface
• Support for various document formats
• Streaming responses with real-time feedback

Technical Architecture 🏗️

Retrieval System 🔍

• Multiple Embedding Models:
  • GTE-large-zh: Optimized for Chinese text understanding
  • BGE-large-zh: Enhanced semantic comprehension
  • BM25: Classical information retrieval algorithm
• Reranking: Uses BGE-reranker-large for context optimization
• Large Language Model: Powered by GPT for natural language generation

How It Works 🛠️

1. Document Processing 📄:

   • Documents are loaded and split into manageable chunks
   • Each chunk is processed through multiple embedding models

2. Query Processing 🔎:

   • User queries are processed through multiple retrieval routes
   • Results are combined and reranked for relevance
   • Most relevant context is selected for the final response

3. Response Generation 💬:

   • Selected context is combined with the user query
   • LLM generates natural and accurate responses
   • Responses are streamed in real-time

Installation & Usage Guide 🚀

Prerequisites 📋

• Python 3.10+
• CUDA-capable GPU (recommended)
• uv package manager (recommended)
• OpenAI API key

Installation Steps 📥

1. Clone Repository

git clone https://github.com/AbyssSkb/Docbot
cd Docbot

2. Install Dependencies

# Using uv (recommended)
uv sync

# Or using pip
pip install -r requirements.txt

3. Environment Setup

• Create a .env file in the project root:

OPENAI_API_KEY=your_api_key
OPENAI_BASE_URL=your_base_url  # Optional
OPENAI_LLM_MODEL=your_preferred_model  # Default: gpt-4o

4. Document Setup

• Create a doc folder in the project root
• Place your documents in the doc folder
• Generate document indexes:

python create_index.py

Running the Application 🏃

streamlit run main.py

Basic Usage 💡

1. Open the provided URL in your web browser
2. Enter your questions in the chat interface
3. View real-time responses based on your documents

Limitations and Considerations ⚠️

1. Language Support 🌐:

   • Primary optimization for Chinese text
   • English support can be enabled by switching to English-language models
   • Consider language-specific requirements for your use case

2. Text Processing 📝:

   • Jieba tokenizer is optimized for Chinese
   • Basic English tokenization support
   • May require adjustment for other languages

3. Document Compatibility 📄:

   • Uses LangChain's DirectoryLoader
   • Some document formats may have compatibility issues
   • Verify support for your specific document types

Contributing 🤝

Contributions are welcome! Please feel free to submit pull requests or create issues for bugs and feature requests.

License ⚖️

This project is licensed under the MIT License - see the LICENSE file for details.