01 · Scan
Point. Tap. Done.
PaddleOCR reads any medical document — prescriptions, lab reports, insurance cards, handwritten notes. Gemma extracts structured data. All on your Snapdragon NPU, no upload.
Point your camera at any prescription, lab report, or clinical note. On-device Gemma reads it, files it, and answers questions about it — without sending a single byte to a server.
Free · No account · Android · Snapdragon NPU
No account required. No waiting for a server. Everything that follows is running on your phone.
01 · Scan
PaddleOCR reads any medical document — prescriptions, lab reports, insurance cards, handwritten notes. Gemma extracts structured data. All on your Snapdragon NPU, no upload.
02 · Vault
Records filed by body system, timeline, medications, and protocols. Browse, search, and edit offline. Your files are plain Markdown — readable forever with or without this app.
03 · Ask
Type or speak any question about your records. EmbedGemma searches your vault semantically and Gemma 4 answers from your own data — no cloud, no latency, no exposure.
The desktop demo walks through scanning a real prescription, asking the AI about drug interactions, and reviewing the Health Vault. Everything you see runs on a $700 Android phone — no server involved.
Every cloud health app has the same seam: a server you have to trust. We removed the seam. Gemma runs on your Snapdragon chip. Records are Markdown files on your own storage. There is no backend to breach — because there is no backend.
80% of health apps share data with third parties (BMJ, 2019). Health Passport collects nothing. Uninstall to delete everything.
Gemma 4 E2B — health Q&A and grounded reasoning. Google AI Edge LiteRT-LM. ~2.6 GB, runs on CPU/GPU.
PaddleOCR — document scanning, 50+ languages. Nexa SDK. Works offline on any Android.
Parakeet-TDT — on-device speech-to-text for clinic transcription. Nexa SDK · NPU.
EmbedGemma — semantic search over your full health vault. Millisecond recall. Nexa SDK · NPU.
OmniNeural-4B — vision analysis for complex documents. Optional 4 GB model via Nexa SDK.
Qualcomm Snapdragon NPU acceleration. GGUF CPU fallback for non-Snapdragon devices. Kotlin + Compose, minSdk 27.
Each of these required submitting a working product, not a pitch deck.