Community & Nonprofit

CareLoop

Care you can trace.

Demo video

What it does

CareLoop is an AI care companion that turns scattered healthcare documents, appointment letters, discharge papers, visit notes, and prescriptions into one continuous, traceable timeline from before the appointment through full recovery. Patients upload whatever documents they already have (PDFs, photos, text). CareLoop extracts structured facts and re-presents them as a unified journey with four phases: Before (plain-language checklist, time-anchored reminders, suggested questions), During (structured visit notes separating actions from info), After (day-by-day recovery plan with medications, restrictions, follow-ups), and Check-Ins (contextual prompts tied to specific warning signs in the patient's own discharge instructions). Every instruction is traceable to its source document. Tap any item to see the exact PDF field it came from. The AI never invents medical content; it only translates, structures, and schedules what clinicians already wrote. Flagging surfaces matches against the patient's own document's warning signs and never a diagnosis, only "your discharge instructions said to watch for this." Built for Priya (34, ADHD, surgery prep), Mr. Osei (71, low health literacy, physiotherapy), Lin (45, non-native speaker, specialist scan), and their caregivers. Accessibility (Light/Dark/High Contrast/Large Text/Simple Mode) is a theme layer over the same structured content, not an afterthought. Grounded. Traceable. Accessible. The care loop that closes.

Inspiration

CareLoop came from a simple problem we kept noticing: after an appointment or procedure, important information can end up scattered across documents, notes, discharge instructions, and things the patient is trying to remember themselves. We wanted to make that whole process easier to follow, especially for people who may already feel overwhelmed by healthcare information. That led us to the idea of one continuous care journey that helps before, during, and after an appointment, while always showing where the information came from.

How we built it

We built CareLoop as a React/Vite frontend with a FastAPI backend and SQLite for local data storage. Gemini is used to extract and organize information from uploaded care documents and to power the grounded CareLoop Copilot. Instead of letting the AI freely generate medical information, we designed the system around structured records, source provenance, citations, and clear separation between clinician-provided information and patient-entered notes. We also focused heavily on accessibility, adding Simple Mode, Large Text, High Contrast, appearance themes, reduced-motion support, contextual check-ins, persistent Copilot chats, and the Before → During → After → Check-ins care flow.

Challenges we ran into

One of our biggest challenges was making the AI useful without allowing it to invent medical information. We spent a lot of time improving grounding, source traceability, citation validation, and safe handling of unsupported medical questions. Another challenge was connecting all the different stages of care so questions prepared before a visit, notes recorded during it, follow-up information, check-ins, and Copilot all felt like parts of the same journey. We also had to balance accessibility and simplicity with showing enough information to actually be useful, and AI processing time was something we had to design the UI around.

What's next

The next step would be turning CareLoop from a hackathon prototype into something that could support real users securely. We would add proper accounts and per-user data ownership, production-grade storage, stronger privacy controls, and integrations with real healthcare systems where appropriate. We would also improve processing speed, notifications and reminders, and expand Copilot while keeping the same grounded, source-first safety rules. Longer term, we see CareLoop becoming a place where a patient can understand and navigate their care history without having to piece everything together themselves.