RAREPATH

Inspiration

The inspiration for RarePath comes from a deeply personal place. My cousin has Duchenne Muscular Dystrophy, and when my family was looking for clinical trials back in Morocco, the experience was incredibly frustrating. Information was scattered across websites, PDFs, and forums, with complex eligibility criteria that took weeks to understand. I realized countless families facing rare diseases go through this same exhausting process, and this hackathon was the perfect opportunity to build something that could make this journey easier for families while helping researchers understand trial landscapes more effectively.

What it does

RarePath is an AI-powered clinical trial intelligence platform that transforms how families and researchers navigate rare disease trials. For families, it uses AI agents to intelligently match patient profiles against live ClinicalTrials.gov data and explains eligibility in plain language. For researchers, it provides sophisticated pattern analysis revealing research trends, institutional networks, and enrollment opportunities. The platform features real-time data enhanced with PubMed research context, creating a comprehensive view that adapts as new information becomes available.

How we built it

I started with Devin to create the initial skeleton and architecture, then transitioned to Windsurf and Claude for the majority of development. The tech stack is React TypeScript with Tailwind CSS, built on Vite. I implemented a multi-agent architecture: SearchAgent for ClinicalTrials.gov data, PubMedAgent for research context, EligibilityAgent for medical criteria evaluation, and ReasoningAgent for human-readable explanations. I integrated real APIs including ClinicalTrials.gov and PubMed, using Vite's proxy for CORS handling. The pattern analysis and knowledge graph features required data processing pipelines to extract insights from large datasets.

Challenges we ran into

My biggest challenge was getting the right amount of PubMed data without overwhelming the system while ensuring research context remained relevant. I also struggled with parsing complex clinical trial eligibility criteria written in technical medical language. Another challenge was building a UI that handles both simple family searches and complex multi-dimensional criteria for medical professionals. Additionally, integrating live APIs while maintaining performance and handling failure scenarios proved more complex than anticipated.

Accomplishments that we're proud of

I am proud of how intuitive the platform turned out, knowing it could help families dozens of research hours while providing clear explanations of complex medical criteria. Technically, I'm particularly proud of my PatternMiningAgent-powered pattern analysis that extracts meaningful insights about research trends and trial landscapes from thousands of data points. The seamless real-time integration of ClinicalTrials.gov and PubMed data, combined with the multi-agent architecture providing reasoned explanations, feels like a genuine step forward in making medical information accessible.

What we learned

This was an incredible learning experience, especially since I'd never used Devin before. Working with multiple AI tools taught me to leverage each one's strengths. I also gained insights into clinical trial data structures and medical API challenges. The project taught me about building applications serving both technical researchers and non-technical families. Most importantly, I learned how AI agents can work together on multi-step problems that would probably overwhelm traditional systems.

What's next for RAREPATH

I envision expanding beyond rare diseases to cover the entire clinical trial landscape, potentially helping millions of patients. I want to add predictive analytics for trial success rates and emerging research identification. Integration with EHR/EMR could enable automatic patient matching. Long-term, we see RarePath becoming the intelligent infrastructure layer for clinical research (connecting patients, families, researchers, and pharmaceutical companies).

Built With

• claude
• clinicaltrials.gov
• devin
• lucide
• pubmed
• react
• tailwind
• typescript
• vite
• windsurf