Caregap

Inspiration

Healthcare access is broken for millions of people who fall through the cracks the uninsured, people in rural areas, non-English speakers, and those who simply don't know where to start when something feels wrong. We kept hearing stories from family and friends who delayed care because they didn't understand their symptoms, couldn't afford a specialist visit just to be told it was nothing, or didn't know which resources were available to them. That gap between "I don't feel right" and "I got the care I needed" is where people get hurt. We wanted to build something that closes it.

What it does

Caregap is an AI-powered healthcare companion that helps users understand their symptoms, find affordable care options nearby, and get guidance in plain language. Users describe what they're experiencing, and Caregap walks them through a conversation that helps narrow down possible causes, flags anything urgent, and points them toward the right next step whether that's a free clinic, a telehealth visit, or immediate emergency care. It's not a replacement for a doctor; it's the friend who helps you figure out whether you need one.

How we built it

We built the frontend with React and Tailwind for a clean, fast interface that works well on mobile since most of our users would be checking it on their phones. The backend runs on Node.js with Express handling the API layer. We integrated Anthropic's Claude for the main conversational reasoning because of its strong handling of nuanced medical language, Gemini for multimodal input (users can upload images of rashes, medication labels, or documents), and OpenAI for specific embedding and classification tasks. Data is stored in a lightweight database, and we deployed the whole thing so it's accessible from any browser.

Challenges we ran into

Getting the AI to be genuinely helpful without overstepping into giving medical diagnoses was the hardest part. We spent a lot of time tuning prompts so the model would ask the right follow-up questions, stay grounded, and always route serious symptoms toward professional care instead of trying to solve them itself. On the technical side, juggling three different AI providers meant dealing with different response formats, rate limits, and latency we had to build a wrapper layer to make it feel seamless to the user. Integrating everything under time pressure while also making the UI look polished was a constant tradeoff.

Accomplishments that we're proud of

We built a working, end-to-end product in 36 hours that actually responds to real symptom descriptions in a way that feels human and safe. The UI is clean, the AI responses are grounded, and the whole flow from "I don't feel good" to "here's what to do next" runs smoothly. We're proud that it handles edge cases gracefully and doesn't hallucinate advice it shouldn't give.

What we learned

We learned how much prompt engineering matters when you're working in a domain as sensitive as healthcare. We also got much better at integrating multiple AI APIs into one coherent experience, and at making design decisions fast without overthinking them. Most of all, we learned that shipping something rough that works beats shipping nothing polished.

What's next for Caregap

We want to expand language support so non-English speakers can use it natively, add a directory of free and sliding-scale clinics with real-time availability, and partner with community health organizations to actually get this in front of the people who need it most. Long-term, we see Caregap as a free, accessible first step into the healthcare system for anyone who's been left out of it.