Medical Prognosis Bot

Inspiration

Wrong prognosis is an issue which has severe consequences. In order to augment the accuracy of doctors as well as lighten their workload, we wanted to create a model which presents possible diagnoses in a reliable manner.

What it does

A user interface based on a chat-bot allows you to write out your symptoms as if you were talking to a doctor. Once you submit, it returns (extremely quickly) the most probably prognosis based on our model.

How we built it

The chat-bot functionality was done through the Groqcloud API. The UI was created with Streamlit and the model was based on xgboost trained on a Kaggle dataset with 41 different diseases.

Challenges we ran into

Debugging presented significant issues near the end with passing in the right value types from the LLM's response to the input required for the model.

Accomplishments that we're proud of

We achieved a 89% accuracy rate with matching or higher recall on a majority of the symptom features we tested. Having done this all in the span of < 24 hours of working, we were proud to have created all the moving parts and facilitated them working together.

What we learned

Next time, we'll spend more time/prioritize the UI development as well as the debugging stage.

What's next for Medical Prognosis Bot

Increases in the dataset the model is trained on to detect a greater array of diseases/symptoms. Making the UI flow better with consistent styling between the chat text input and output.