// the find
LifeValue/HealthWallet.me
Open-source, patient-controlled health record app with on-device AI. Aggregates medical data from 52K+ providers via FHIR R4. Offline-first. Flutter.
HealthWallet.me is a Flutter app that aggregates personal medical records from US healthcare providers via FHIR R4, stores everything locally on-device, and runs a quantized LLM (llama.cpp) to extract structured data from scanned paper documents. It's aimed at privacy-conscious patients who want a single view of their health history without trusting a cloud service with their medical data. The backend aggregation relies on a self-hosted FastenHealth instance, so this is a two-component setup, not a standalone app.
The on-device LLM pipeline is the genuinely interesting part — running Qwen3-VL or MedGemma locally via llama.cpp to OCR and structure medical documents means no PHI leaves the device, which is a real architectural commitment, not just a marketing claim. The proximity sharing implementation (cross-platform iOS/Android, view-only receiver, screenshots blocked) is a thoughtful solution to a real problem — handing records to an ER doctor without cloud dependency. The architecture is clean Flutter: BLoC, GetIt, AutoRoute, Drift, code-gen heavy — a developer joining this project won't spend a week figuring out where state lives. The Drift schema versioning (9 tracked schema snapshots) shows the data layer is being maintained seriously rather than treated as an afterthought.
65 stars and 2 forks after shipping to both app stores is a signal that adoption is thin; with medical data on the line, a small contributor pool means security issues could go unnoticed for a long time. The self-hosted FastenHealth backend is a real barrier — non-technical patients (the stated audience) won't spin up Docker to connect their Epic records, and the README doesn't mention any hosted option. Android CPU-only inference on a 2.5 GB model is going to be painful on anything but flagship hardware; the RAM requirements table (12 GB for full Android support) quietly excludes most of the Android market. The 'AI health insights' and 'AI Note taking' roadmap items are vague, and given that this is already processing actual medical records, there's no mention of how incorrect AI extractions are surfaced or corrected — demo correction templates exist in the assets but there's no documentation on when or how they're used.