ViMedicine: An Vietnamese LLMs agent for drug and illness suggestion

Introduction

In this project, we present a medicine agent for Vietnamese medical suggestion. The medicine agent can:

Predict illness based on patient's symptoms
Suggest medicine based on disease
Check specific medicine's suitability based on disease and patient's symptoms
Predict medicines' side effects or interact or violate patient's medical record

We use OpenAI GPT-3.5 model to read documents and generate medical entities, but the agent flow is not restricted to OpenAI GPT model or to Vietnamese medical domain.

⚠️ Notes: The full document is comming soon.

Install requirement

Input your chat model key in config/.env file. Install via pip:

pip install -r <requirements.txt>

Initualize Milvus server and deploy API

Initualize milvus server and create collection

docker-compose up -d
python setup_milvus.py

Notes: for the data, check out Here

Deploy API server

uvicorn main:app  --reload --port=<port> --workers=<n_worker>

Acknowledge

This is a research project, so we do not guarantee the accuracy of the model. Since the performance heavily depends on database (in vector store) quality, the current model precision might not meet the expectation for production use.

Citation

@misc{vimedicine,
  author = {Tran Tuan Anh, Phan Quang Hung, Dung Manh Nguyen},
  title = {ViMedicine: An Vietnamese LLMs agent for drug and illness suggestion},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/nmd2k/vi-medicine/}},
}

Our team:

Tran Tuan Anh*
Phan Quang Hung*
Dung Manh Nguyen*