Plans4You | Devpost

Inspiration

The idea for Plans4You came from a common and frustrating experience: navigating the complex world of health insurance. Many users are faced with dozens — sometimes hundreds — of plans, each filled with confusing terminology like premiums, deductibles, copays, coinsurance, provider networks, and more. These details are often buried in long documents and hard to compare side-by-side.

The time and effort required to research these plans can be overwhelming, and making the wrong choice can lead to high out-of-pocket costs or limited access to care. For example, some users may choose a plan with the lowest premium, only to realize later that their regular doctor isn’t in-network or that frequent services cost more than expected.

Plans4You aims to solve this by guiding users through the process using intelligent filtering and AI-based plan review. Our goal is to empower people with personalized, easy-to-understand recommendations so they can make informed decisions with confidence.

What it does

Plans4You helps users discover the most suitable health insurance plans based on their personal needs. It uses publicly available healthcare datasets and combines them with user input to filter and rank plans. A agentic AI system then reviews the top results and identifies one plan as the “BEST MATCH,” providing an explanation to help users understand why.

User Input: Users provide key information such as age, income, dependents, and whether they need dental coverage through a simple, web-based form.
Data Analysis: The system uses CMS (Centers for Medicare & Medicaid Services) public datasets to filter plans based on user preferences.
Agent Scoring: A insurance decision-making agent evaluates the plan features and simulate how a real advisor might weigh trade-offs.
BEST MATCH Review: The AI agent uses NLP to qualitatively review the top plans and selects the one that best fits the user's profile.
Results Display: The top plans are shown on the frontend, with the BEST MATCH clearly highlighted and accompanied by a short explanation.

How we built it

Data Processing: We used Python scripts and MongoDB Atlas to parse, clean, and organize multiple CMS datasets, including Plan Attributes PUF and Rate PUF, which contain millions of records.

AI Recommendation System: The backend includes a agent system developed in Python, where each agent handles a specific domain (insurance and decision-making). These agent evaluates the top plans based on user preferences.

Frontend: We built the user interface with Next.js to ensure responsiveness and ease of use.

Backend: Flask is used to handle API endpoints, data requests, and communication between the frontend, database, and AI modules.

Challenges we ran into

Working with large datasets: CMS datasets are extensive and sometimes inconsistent. Merging them correctly and efficiently required careful data handling and optimization to avoid storage issues.

Bridging the language gap: Translating complex insurance terms into plain, understandable language was essential to make the app truly user-friendly.

Building trust in AI recommendations: We wanted users to feel confident in the plan suggestions, so we focused on limiting hallucinations and ensuring explanations were concise and accurate.

Maintaining accuracy: Because this involves healthcare decisions, we had to be very careful with how the AI interpreted and summarized plan details.

Accomplishments that we're proud of

Successfully filtered and analyzed over one million rows of health plan data.
Designed a functioning MVP that generates real-time plan recommendations.
Developed a agentic reasoning system that simulates real-world decision-making.
Built a responsive frontend and integrated it seamlessly with the backend and AI modules.
Created a simple, transparent interface for a complex and sensitive topic like health insurance.

What we learned

Through building Plans4You, we gained hands-on experience working with real-world, large-scale healthcare datasets. Parsing and integrating data from multiple CMS sources taught us how to handle inconsistencies, optimize performance, and design scalable data pipelines. We also deepened our understanding of how to apply AI in a way that goes beyond basic scoring — by a agentic system that simulates different perspectives and reasons through trade-offs to provide more thoughtful, user-aligned recommendations.

We also learned the importance of designing for trust and usability, especially in sensitive domains like healthcare. It wasn’t enough for the AI to be accurate — it had to be understandable and transparent. We spent time refining how plan details and recommendations were communicated, ensuring the interface was clear and the reasoning behind each suggestion was easy to follow. This balance between technical complexity and user-centered design was one of the most valuable lessons we took away from the project.

What's next for Plans4You

Next, we plan to expand coverage to include dental, vision, and prescription drug plans. We also want to improve our AI agent’s ability to explain recommendations in plain language and integrate real user feedback to guide future iterations. Finally, we’re working on hosting the website and building a mobile version of Plans4You to make the tool more accessible and convenient for users on the go.