MedChat won the Cohere RAG Hackathon held at the University of Waterloo! News article.
MedChat serves a dual purpose:
-
Querying Medical Literature: Using RAG om medical literature, MedChat allows easy interaction with medical literature, aiding healthcare providers in keeping up-to-date with the latest research findings, treatment protocols, and clinical guidelines.
-
AI-Powered Diagnosis Assistance: Beyond literature, MedChat integrates pre-trained medical models that assist in disease diagnosis. Doctors can submit MRI scans, X-rays and possibly other data, and MedChat can call other models to extract meaningful insights and diagnoses, offering a second opinion in medical diagnoses.
-
Utilizing Cohere API: The backbone of MedChat is the Cohere API, including Cohere Chat for natural language processing, Classify for classifying user intent, and Rerank for optimizing response relevance.
-
Frontend Development with Streamlit: We chose Streamlit for its simplicity.
-
Integrating Tensorflow for Running Medical Models: The core functionality of disease diagnosis is powered by medical models run on Tensorflow. This integration allows MedChat to process complex medical data and provide accurate, AI-driven diagnostic suggestions.
A primary challenge in developing MedChat was:
- Detecting User Intent Accurately: To ensure that MedChat responds appropriately to the queries of medical professionals, a critical task was to accurately detect user intent. This involved discerning whether a query was seeking medical literature or a diagnosis, and then triggering the correct functions within the application. We did this using Cohere Classify, which is not scalable. Now that Cohere API supports Function calling, that would be a much better tool.
