Hybrid RAG System

This project is a Retrieval-Augmented Generation (RAG) system implemented using Python, LangChain, and the DeepSeek R1 model. It combines traditional retrieval techniques (BM25) with modern dense embeddings (FAISS) to build a highly efficient document retrieval and question-answering system.

---

Features

Current Features

1. Hybrid Retrieval: Combines BM25 and FAISS for robust and accurate document retrieval.
2. Multi-PDF Support: Users can upload multiple PDF files for processing, which are stored in a dedicated data/ directory.
3. Streamlit UI: A user-friendly interface for uploading files, asking questions, and viewing results.
4. Tracing and Analytics: Integrated tracing with LangSmith to analyze performance and monitor usage.
5. Custom LLM Integration: Uses the DeepSeek R1 model (via Ollama) for question answering.
6. Dynamic Context Handling: Automatically handles and prepares context for queries.

Planned Features

Enhanced Memory Features

1. Context-Aware Memory: Implement dynamic context retention to remember the flow of conversations.
   • Framework: LangChain's updated ConversationBufferMemory or EntityMemory.
   • Use Case: Retaining context across multiple user queries for more coherent interactions.
2. User-Specific Memory: Allow memory reset or persistence for different users.
   • Framework: Redis or PostgreSQL for memory persistence across sessions.

File Management

1. Support for Multiple File Formats:
   • Additional Formats: Microsoft Word, CSV, and image files (via Tesseract or Amazon Textract for OCR).
   • Library: python-docx for Word, pandas for CSV, and pytesseract or AWS Textract for images.
2. File Listing UI:
   • Feature: A sidebar UI for managing uploaded files (view/delete).
   • Library: Streamlit components (st.sidebar and st.selectbox).

Advanced Tracing and Analytics

1. Usage Analytics:
   • Framework: LangSmith or OpenTelemetry.
   • Metrics: Number of queries, response times, and user feedback.
2. Error Logging:
   • Framework: Sentry or Python's built-in logging library.
   • Storage: Centralized logs for troubleshooting.

API Integration

1. REST API:
   • Framework: FastAPI for building a RESTful API.
   • Use Case: Exposing functionalities for external applications.

Security Enhancements

1. Access Control:
   • Framework: FastAPI Users for authentication and role-based access.
   • Feature: Secure endpoints for API access.
2. Data Encryption:
   • Library: cryptography for encrypting files and query results.

Enhanced Retrieval

1. Advanced Retrieval Techniques:
   • Framework: DPR (Dense Passage Retrieval) using Hugging Face models.
   • Improvement: Replace FAISS with Weaviate or Milvus for better vector storage and search.
2. Semantic Clustering:
   • Library: Scikit-learn for clustering similar documents.

Performance Optimization

1. Parallel Processing:
   • Framework: concurrent.futures or multiprocessing.
   • Use Case: Faster processing of large files.
2. Caching System:
   • Library: Redis or Memcached for caching embeddings and document chunks.

Feedback System

1. User Feedback Loop:
   • Framework: Streamlit widgets for rating responses.
   • Use Case: Improve system accuracy with user feedback.
2. Interactive Debugging:
   • Feature: Flag incorrect answers directly from the UI.

Testing and CI/CD

1. Automated Testing:
   • Framework: Pytest for unit and integration tests.
   • CI Tool: GitHub Actions for continuous integration.
2. Continuous Deployment:
   • Tool: Docker and AWS CodePipeline for seamless updates.

Community Engagement

1. Knowledge Base:
   • Platform: GitBook or ReadTheDocs for user and developer documentation.
   • Content: Guides, FAQs, and tutorials.
2. Open Source Contribution:
   • Platform: GitHub for hosting and collaboration.
   • Feature: Contributor guidelines and issues for community involvement.

---

Installation

Prerequisites

• Python 3.9+
• Pip
• Ollama installed (installation guide).

Steps

1. Clone the repository:

   git clone https://github.com/your-repo/hybrid-rag.git
   cd hybrid-rag

2. Create and activate a virtual environment:

   python -m venv rag_env
   source rag_env/bin/activate  # On Windows: rag_env\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Install Ollama and pull the DeepSeek R1 model:

   # Install Ollama
   brew install ollama  # macOS
   
   # Pull the DeepSeek R1 model
   ollama pull deepseek-r1:1.5b

5. Run the Streamlit app:

   streamlit run app.py

---

Usage

1. Upload one or more PDF files via the Streamlit UI.
2. Ask questions based on the uploaded documents.
3. View responses and source documents.

---

Contributing

We welcome contributions! Please check the Contributing Guidelines.

---

License

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

---

References

• LangChain Documentation
• DeepSeek R1 Model
• Streamlit Documentation
• FAISS
• BM25