Referra

Inspiration

Our inspiration for this project was our teams prior experience in building a referral system for US services for people in need.

What it does

Referra is an app that allows Indian patients or providers to find and discover facilities near them that will best serve them. Our main screen is connected to a genie chatbot which has action to a random forest ml model we trained to provide the most relevant locations to our users. Factoring in their age and gender by correlating facilities to nfhs data, as well as any conditions/services users are looking for, we are able to determine what serves them best. Coordinators also have the option to manually peruse available facilities, if they are not pleased with the offerings provided.

How we built it

Referra was built ground up on top of Databricks. We utilized a medallion architecture to clean and refine the data. We utilized AppKit and Databricks Apps for a complete Application deployment. Genie (through the conversational API) was utilized for our chat functionality within the application. LakeBase was used with synced tables for low latency reads and was used to track usage anayltics on the app. We used ML Flow to serve and random forest ML Model which is utilized through Genie. This model uses the users age, location, gender, and the available resources at the facility (including equipment, capabilities, capacity, number of doctors etc.) with sufficient weights applied to each section so that the most relevant results are given to each user.

Challenges we ran into

The primary challenge our team focused was limits/queuing related to the free edition. We ended up not being able to sync all of our delta table to the Lakebase instance since the free edition did not give us priority for our sync pipelines to run within a normal time. (Often our sync pipelines were stuck in the "Creating..." state for ~45 minutes before executing).