Patient Facing - Facility Finder

Inspiration

We were looking to create a patient or care coordinator facing interface that would allow quick matching to relevant facilities within 50km of the patient's general location.

What it does

The app will show a map of India, when prompted in the chat with the description of a patient, their needs, and at least a general location, it will find matching facilities within 50kms and display a data quality score to express the level of confidence that the data is accurate and up to date.

How we built it

We used the supplied data sets and synthesized the facility specific into a singular content column (service_description). A vector index was created on this field along with searchable fields for latitude and longitude of the facility. When the user input is submitted, it will assign coordinates to the patient location and look in an approximate 50 km radius to pull all facilities in the area. From there, the vector index allows the top 20 likely matches to pass to the LLM that will pick 3 that appear to be the best fit.

Challenges we ran into

Numerous. Most of them personal. We had not worked with integrating full agents into an app framework yet, only direct calls to LLM endpoints. The authentications and input/response formatting took a minute.

Accomplishments that we're proud of

Figuring out the input/response formatting.

What we learned

Better use of Code Genie. "Talk through it first, code second"

What's next for Patient Facing - Facility Finder

It was a decent project for us in terms of experience and learning the tools at hand better. We work for a healthcare provider that could have use of similar solutions.

Built With

• databricks
• python