Le ChangeHealth

Finally understand your health insurance policies. Decodes complex insurance guidelines into clear info, empowering patients to make informed decisions and navigate their healthcare effortlessly.

Comment

Inspiration

Navigating insurance coverage for medications and medical procedures is a daunting challenge for millions of Americans. The complexity and opacity of health plan systems leave many struggling to understand their eligibility and rights, despite insurers' claims of transparency. Often, the detailed policy documents provided are practically inscrutable to the average person.

What it does

To empower patients and demystify this process, we introduce a pioneering solution: leveraging the capabilities of Mistral alongside advanced Large Language Models (LLMs). Our technology delves into the dense thicket of coverage policies, decoding and simplifying them into clear, actionable information. This task, virtually impossible for individuals to tackle, is accomplished in mere seconds by our system.

By bridging the gap between complex insurance documentation and patient understanding, we're not just providing a service; we're restoring power to the patients, enabling informed decisions about their health care.

How we built it

1) Get data (clinical policies)

  • Scrape from insurance website
  • Label with drug/procedure type + insurance type

2) Embedding

  • Embed policy
  • Attach metadata to each embedding

3) Create a table to store information

  • Chunk id
  • Embedding vector
  • Drug/procedure type
  • Insurance type
  • Document version

4) Query and filter by metadata

  • Extract from user query the drug/procedure type + insurance type
  • Embed user query
  • Based on user query labels, parse table and assess cosine similarity
  • Retrieval

5) Deployment

  • Database
  • Backend
  • Frontend

Challenges we ran into

Main problem: complex metadata attachement to embedding not supported by Faiss. Tried multiple other vectorDB which none of them were flexible enough for our work. Ended up manually creating the entire process with postgresql.

Accomplishments that we're proud of

Being able to actually compare different insurance plans, based on the most hidden documents that are actually relevant to the patient. Not just the co-pay, but am I going to qualify for Ozempic this month?

What we learned

Everyone in the team stepped out of their comfort zone to build this life-changing product. The main learnings were human, working hand-in-hand for a bigger purpose.

What's next for Le ChangeHealth

1) Populate DB with more data:

  • Other drugs
  • Procedures

2) Cover more insurance guidelines (further than Kaiser, Anthem, Cigna, Aetna)

3) Track change of insurance guidelines over time

  • Patient “subscribes” to a drug/procedure insurance companies
  • Alert patient whenever the drug/procedure changes policy in their insurance and other insurances

Join us in transforming the future of health insurance navigation, making it transparent, accessible, and patient-centered!

Built With

Share this project:

Updates