The goal is to create a healthcare diagnosis chatbot that assists users by analyzing their symptoms, utilizing databases that contain real doctor-patient conversations. The solution involves leveraging task-specific datasets, Retrieval-Augmented Generation (RAG) frameworks, vector databases, and quantized LLM fine-tuning (QLoRA). The system aims to improve LLM accuracy by providing it with external accurate data, based on the user's needs and to reduce LLM hallucinations via finetuning.
- Python: Primary language used for backend
- LangChain: To generate templates for few-shot prompting
- PyTorch: Primary ML framework used
- bitsandbytes: For QLoRA fine-tuning
To set-up the project, follow the below commands:
- Run
python install -r requirement.txt
from LLM_model folder. - Run
python3 main.py
from LLM_model folder to spawn the server for interacting with LLM model. - Run
npm run
from frontend folder to start the frontend.
To test the backend using cli:
- Run
python install -r requirement.txt
from LLM_model folder. - Run
python3 test_prompt_executor.py
from LLM_model folder to start the cli tool for interacting with LLM model.