KomeKare is an AI-powered family caregiver coordination app built natively with Expo and Firebase at the Divergent Hackathon, where it placed 3rd. It links caregivers, family members, and healthcare providers around a shared patient profile, using Google's Gemini 1.5 Flash model to summarize care data into insights each stakeholder can actually act on.
Caregiving is often fragmented across family members, professional caregivers, and medical providers who lack a shared view of the patient. Critical information gets lost in text threads, paper notes, and verbal handoffs — leading to missed medications, duplicated tasks, and poor outcomes.
Native Expo app with Firebase as the backend
Built as a native React Native app with Expo, backed entirely by Firebase — Auth for identity, Firestore and the Realtime Database for care data, and Storage for uploads.
Reason:
A 48-hour hackathon timeline meant we needed a backend that required zero server setup, had generous free-tier limits, and gave us real-time data sync between caregivers, family, and providers out of the box.
Tradeoff:
Firebase's NoSQL model and vendor lock-in trade off long-term data-modeling flexibility for the speed of shipping a working real-time app in a weekend.
Zustand with AsyncStorage-backed persistence
Client state (current patient profile, session, cached care data) is managed with Zustand, using a persistence middleware bound to AsyncStorage so state survives app restarts.
Reason:
Zustand's minimal boilerplate let us wire up shared state across screens quickly without the ceremony of Redux, while still getting durable local persistence for free.
Tradeoff:
Less structure than a more opinionated state library, which requires more discipline as the app grows to avoid ad-hoc state shapes.
Gemini 1.5 Flash for AI care insights
Health logs, medication schedules, and caregiver notes are summarized by Google Gemini 1.5 Flash into natural-language insights surfaced to each stakeholder.
Reason:
Different stakeholders need different views of the same data — a caregiver needs a quick task list, a family member needs a plain-language update, a provider needs a clinical summary. Flash gave us low-latency summarization that fit a live demo.
Tradeoff:
LLM responses can hallucinate, so care-related summaries need careful prompting and framing as assistive, not authoritative, especially under hackathon time pressure to harden that further.
- Won 3rd place at the Divergent Hackathon.
- Built a working native Expo/Firebase prototype in under 48 hours.
- Demonstrated Gemini-generated care summaries tailored to caregiver, family, and provider stakeholder views.
- Hackathon speed requires ruthless prioritization of the demo-able core.
- Healthcare AI requires extra attention to accuracy — hallucinations are unacceptable in care contexts.
- Coordination problems are often better solved by better information flow than new features.