Dense Retrieval-based Medical Chatbot with T5 and Redis Stack

This project is a Vietnamese-language chatbot system designed for medical question answering. It uses a Retrieval-Augmented Generation (RAG) architecture with the following components:

• VietAI/vit5-base: Fine-tuned with LoRA for text generation.
• Sentence Transformers (vinai/phobert-base): For embedding titles, questions, context.
• Redis Stack with HNSW (Hierarchical Navigable Small World): Vector database for semantic search.
• FastAPI: Lightweight backend API.
• Docker: Deploy the entire system (Redis Stack + FastAPI backend) in containers using docker-compose.
• Data Source: Medical knowledge is derived from public content on YouMed.vn.
  • Note:
    • The raw crawled dataset is not provided due to data protection policies.
    • The data collection and preprocessing pipeline is not included in this repository.
• Trained Model: baduyne/vnt5-medical-gqa
• Evaluate Metric: BLEU = 41.4

---

Features

• Retrieve relevant medical context from a structured knowledge base
• Answer natural language questions using a fine-tuned Vietnamese language model
• Real-time semantic search powered by Redis + HNSW
• Ready-to-deploy with Docker

---

Architecture Overview

Getting Started

1. Clone the Repository

git clone https://github.com/baduyne/Medical-Consultant.git
cd Medical-Consultant

2. Redis Stack + App

docker-compose up --build
docker run --gpus all -p 8000:8000 your_image_name

# Set up nvidia-container-toolkit:
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker

---

Credits

• VietAI for vit5-base
• YouMed.vn for medical content
• HuggingFace, Redis Stack, SentenceTransformers

---