This repository hosts the code and scripts for developing an empathetic medical chatbot, a project led during my time as an ML Researcher at Johns Hopkins University. The chatbot is based on the LlaMA2 model and is designed to improve patient interactions by providing empathetic and accurate medical responses.

• Accuracy Improvement: Enhanced the chatbot's response accuracy to 88.7% on a human-annotated test dataset.
• Innovative Fine-Tuning: Implemented Direct Preference Optimization (DPO) and Reinforcement Learning from Human Feedback (RLHF), specifically Fine-Grained RLHF, to fine-tune chatbot responses for improved empathy, factuality, and quality.

• training/: Contains scripts and notebooks for fine-tuning the LlaMA2 model with conversational data, Includes scripts for instruction tuning to enhance the model's empathy and accuracy.