Skip to content

README.md

Latest commit

 

History

History
68 lines (47 loc) · 2.85 KB

README.md

File metadata and controls

68 lines (47 loc) · 2.85 KB
Raw

Medical RAG System (Single-Model Architecture)

This project is a high-performance Retrieval Augmented Generation (RAG) application optimized for the medical domain. It utilizes a Single-Model Architecture to deliver precise, context-aware medical answers directly in Vietnamese, eliminating the need for intermediate translation layers and significantly reducing latency.

🚀 New Architecture: "Direct Vietnamese Processing"

By leveraging state-of-the-art Large Language Models (LLMs) with strong native support for Vietnamese, the system bypasses the traditional "translation bridge" approach (Vi -> En -> Vi). This results in a cleaner pipeline and more natural language generation.

Simplified 3-Stage Workflow:

  1. Retrieval: Advanced semantic search using BGE-M3 to locate relevant medical documents.
  2. Reasoning: Gpt-oss 20b bnb 4bit analyzes the retrieved context and performs medical reasoning directly in Vietnamese.
  3. Response: Generation of evidence-based answers with strict source citation.

🧠 Core Models

  1. Medical Logic & Reasoning:

    • unsloth/gpt-oss-20b-bnb-4bit (Current): A powerful 20B parameter model optimized for 4-bit quantization, offering superior reasoning capabilities while fitting within 16GB VRAM.

  2. Embedding: BAAI/bge-m3

    • Retained for its State-of-the-Art multimedia and multilingual retrieval performance.

🖥️ System Interface

Below are screenshots of the running system:

1. Login Screen Secure access via predefined credentials (admin / 123456). Login Interface

2. Workspace (Chat Interface) The primary interface for medical professionals to query the knowledge base. Workspace Interface

📦 Installation & Usage

1. Requirements

  • Python: 3.10+
  • Hardware: NVIDIA GPU with CUDA support (Minimum 16GB VRAM to load the 20B model).

2. Install Dependencies

pip install -r requirements.txt

3. Data Ingestion

Place your medical PDF documents into the Medical_documents/ directory and run:

python ingest.py

4. Launch Application

python app.py
  • Access the Web UI at: http://localhost:7860

📂 Project Structure

  • app.py: Main application logic (Single-Model RAG Pipeline).
  • ingest.py: Document processing and vector ingestion script.
  • Medical_documents/: Directory for input PDF files.
  • chroma_db/: Vector database storage (ChromaDB).
  • Images/: Interface screenshots.

Medical Disclaimer: This AI system is for informational and reference purposes only. It is not intended to replace professional medical diagnosis, advice, or treatment. Always consult with a qualified healthcare provider.