CareML

Inspiration

The inspiration for CareML came from the need to empower clinicians with a tool that allows them to easily analyze clinical data and generate insights without requiring deep technical expertise. The idea was to bridge the gap between complex data analysis and healthcare professionals, making it easier to visualize and understand patient data for better decision-making.

What it does

CareML enables clinicians to interact with a clinical database through a natural language interface. Users can query the database using simple language, and the tool automatically converts these queries into SQL. It also provides visualizations of the data, allowing clinicians to see trends and statistics at a glance. We also added another feature which allows users to implement a model on their dataset that we pretrained and picked for them from multiple AI models.

How we built it

The tools we used to build CareML are python, nextJS, PostgreSQL, LLaMA, Tembo, pgAdmin, flask, sklearn.

Accomplishments that we're proud of

We’re proud of creating a functional prototype that allows users to interact with clinical data using natural language. The integration of LLaMA to generate SQL queries from natural language was a significant achievement, as was developing a system that can visualize the data in meaningful ways. We're also proud of automating the machine process by creating a pipeline that preprocesses the data (encoding, scaling) and trains multiple models to choose the best one and provide it to the user.

What we learned

Throughout this project, we learned a lot about the challenges of natural language processing and database management. We also gained experience in integrating different technologies and handling real-world data issues, such as schema mismatches and data formatting. Additionally, we learnt about machine learning techniques and redundancies.

What's next for CareML

In the future, we plan to expand CareML by adding more advanced data analysis features, improving the accuracy of SQL query generation, and integrating with more complex datasets. We also aim to enhance the user interface to make it even more intuitive for clinicians to use. We also intend to have a higher accuracy rate on predictions while building the trained models.