CareCompass

Inspiration

The first week after a cancer report is a literacy problem as much as a medical one. Families are handed pathology language, receptor lines, and scan impressions, then asked to be "good historians" at the next visit. I wanted a tool that sits at the kitchen table: calm, honest about what it cannot do, and useful before the appointment starts.

What it does

CareCompass has three connected flows.

1. Paste or upload a pathology report, imaging report, or visit summary. The app explains it in plain language at Everyday, Classroom, or Clinical-plain reading levels, with a glossary and flags for questions to ask the care team.
2. It builds a prioritized list of visit questions the user can edit and export or print as a one-page PDF.
3. A symptom and side-effect journal stores daily check-ins and a printable snapshot. Red-flag guidance (for example, fever during cancer treatment) is drawn from public NCI and ACS pages and cited in the app.

It never diagnoses, never gives treatment advice, and never predicts survival. A disclaimer stays visible. Health notes stay in the browser. There is no account.

How we built it

Next.js, TypeScript, Tailwind CSS, and shadcn/ui, ready for Vercel. Report text and journal entries use localStorage. PDF text is extracted in the browser. Demo mode ships with fictional sample reports and canned explanations so the product works without an API key. A live OpenAI-compatible provider can be added with LLM_API_KEY, LLM_BASE_URL, and LLM_MODEL. Safety checks run before any explanation is returned. Automated tests cover the disclaimer and the refusal rules.

Challenges we ran into

Writing explanations that feel specific without becoming a second medical opinion. Keeping red-flag copy faithful to public sources without inventing numbers. Designing a UI that feels trustworthy on a phone in a waiting room, not like a startup dashboard.

Accomplishments that we are proud of

A complete demo path judges can click with no credentials. Safety refusals that still leave the person with a next step: take the question to clinic. A one-page print sheet that looks like something a family would actually fold into a folder.

What we learned

Healthcare AI is mostly a product-boundaries problem. The useful work is translation, memory, and preparation. The harmful work is certainty. Citing NCI and ACS in the journal made the red flags feel like a map back to reputable pages, not like the app inventing triage.

What's next

Clinic-branded after-hours cards a care team could issue. A shared print packet that combines the explanation, questions, and journal on two pages. Optional on-device models so live mode never has to leave the laptop.

Built With

• localstorage
• nextjs
• openai
• shadcn
• tailwindcss
• typescript
• vercel