BillBuster

• 

  Medical procedure search page with autocomplete suggestions

• GIF

  Medical procedure search and compare workflow

• 

  Medical procedure compare page with pricing comparisons across hospitals in US

• 

  Medical bill upload page that generates negotiation script

• GIF

  Medical bill upload workflow

• 

  Medical bill form page that generates negotiation script

• GIF

  Medical bill form workflow

• 

  Landing page

Inspiration

I got hit by a car walking to campus in college and the medical bill was what freaked me out the most. When I first read about the Hospital Price Transparency Initiative via a Postman Engineering Blog Post - How APIs Can Help with US Healthcare Price Transparency, I have always wanted to spin out a project that would've been able to help put minds at ease when it came to healthcare related debt and bills.

What it does

1. Generate medical bill negotiation scripts to help lower prices via a form or PDF / PNG file upload.
2. Compare prices for common medical procedures across hospitals in the United States.
3. Search functionality of common medical procedures with autocomplete suggestions.
4. Integrates Perplexity's Sonar API using sonar-reasoning-pro & sonar search models for dynamic generation of data in unstructured and structured responses.

How we built it

• Frontend: React, Material-UI, React-Query, remark-gfm, Vite
• Backend: Koa, DuckDB, Perplexity's Sonar API
• Tooling: pnpm, nvm, Typescript
• Data Sources: CMS Hospital Data, Hugging Face - atta00/cpt-hcpcs-codes

Challenges we ran into

• Data Quality around available hospitals and medical procedures. I wanted to find better data to be used in parts of my application such as the search suggestions of medical procedures to help augment context into some prompts.
• Inconsistent JSON Schema responses for specific hospital procedures from Sonar API. I believe this is due to how there are a number of hospitals non-compliant with the Hospital Price Transparency Initiative or the majority compliant hospitals using MyCharts by Epic Systems which utilize a reCAPTCHA

Accomplishments that we're proud of

• Utilizing Perplexity's Sonar API request options such as specifying domains, using user base64 encoded files, and returning JSON Schema response to have some structured responses that can be utilized in a dynamic yet structured way.
• Learning about the progress in data quality and compliance of hospitals since my last attempt 2 years ago -- working on a side project related to hospital transparency pricing, can be attributed to executive order.
• Finishing the side project on my own with a more mature perspective coming from my current role -- use of pnpm and completion of similar features.

What we learned

• Ease of use when it comes to integrating external API's that use generative AI. Particularly the use of JSON Schemas to output data in a structured way.
• The varying options that can be applied to large language models to generate a response to questions when it comes to prompt engineering depending on the use cases.

What's next for BillBuster

• Extracting machine readable files of medical procedures from local hospitals, normalizing data and loading onto a database to provide a better experience and improve the quality of the data.
• Introducing a map based visualization via mapbox-gl to be able to display user localized data.
• Providing enhanced filtering of medical procedure data using Elasticsearch.

Built With

• koa
• pnpm
• react
• typescript