Inspiration
We were inspired by our combined group interested. Two of us work in insurance and one in healthcare, so we thought creating an agent for assisting patients to find their best insurance policy was a perfect intersection.
What it does
Berries and Cream is an intelligent assistant that helps patients compare and choose Medicare prescription drug plans. It evaluates plans based on:
- State-specific availability
- Monthly drug premiums
- Annual deductibles
- Low-income subsidy eligibility
- Gap coverage availability
- Total premium breakdowns
It allows for guided or natural language queries, empowering patients to make informed, personalized choices.
How we built it
- Databricks SQL: We created a temporary view combining multiple Medicare plan datasets using SQL joins and transformations.
- Data Cleaning: We normalized messy real-world data (e.g., “x”, “Yes”, blanks) and ensured only complete entries were used.
- LangChain Integration: We used
langchainanddatabricks-sql-connectorto create a conversational agent capable of querying the cleaned dataset. - LLM Agent: The agent supports intelligent plan recommendations via Databricks-hosted LLMs.
Challenges we ran into
Accomplishments that we're proud of
- Seamlessly combining SQL data transformation with LLM-based question answering
- Building a prototype that could easily be extended to other healthcare use cases
- Making Medicare data accessible through natural language
- Creating a truly collaborative product across insurance and healthcare backgrounds
What we learned
- How to preprocess and join large real-world healthcare datasets
- How to build and connect LangChain agents with Databricks SQL
- The importance of simplifying plan details for better patient understanding
- Practical strategies for integrating LLMs into real-time decision support tools
What's next for Berries and Cream
- Building a user-friendly web interface (e.g., Streamlit or React)
- Incorporating patient-specific preferences and subsidy eligibility
- Matching plans to specific medications using drug-level databases (e.g., RxNorm)
- Expanding to include Medicare Advantage, dental, or vision plan options
- Evaluating real-world impact with pilot users or provider feedback
- For team bonding, getting cookies together in Florida.
Log in or sign up for Devpost to join the conversation.