family health records / privacy-first mobile AI
MediYaad
a privacy-first mobile record organizer that helps Indian family caregivers turn scattered medical documents into a verified, shareable family health timeline.
overview
MediYaad is a privacy-first mobile record organizer for Indian family caregivers. The idea came from a simple problem: prescriptions, lab reports, scans, and medication histories often live across paper folders, WhatsApp chats, and different family members. When a doctor asks what changed or which medicine appeared on the latest prescription, the information may exist, but it is difficult to find, organize, and trust.
I built a production-oriented vertical slice using Expo, React Native, TypeScript, Supabase, and PostgreSQL. It includes phone sign-in, household consent, private multi-page uploads, queued extraction, evidence-by-evidence review, a verified family timeline, medication reconciliation, an offline emergency card, and server-rendered Doctor Packs. Core flows are localized in English, Hindi, and Telugu, while original medical text remains untouched and linked to its source page.
The most important design decision was to keep AI in a draft role. MediYaad does not diagnose, recommend treatment, or silently decide that a medicine is current. Extracted fields must be reviewed before they become verified records. PostgreSQL row-level security, private storage, short-lived URLs, audited RPCs, SQLCipher-backed offline storage, explicit consent, and controlled deletion make privacy part of the architecture instead of a final checklist.
why it matters
MediYaad is one of my most complete health-product builds because it connects product design, mobile engineering, AI extraction, privacy, localization, and clinical caution in one system. The goal is not to replace a doctor or build another complicated hospital record. It is to help a family arrive with clearer, verified information when it matters.
what i built
- built a secure Expo and React Native application for organizing family medical records across multiple profiles.
- designed a review-first extraction workflow where AI output remains a draft until every field is confirmed or corrected.
- implemented private multi-page uploads, durable queue processing, retries, evidence links, and explicit failure states.
- created a verified health timeline and medication reconciliation flow that distinguishes a document mention from a confirmed current medicine.
- generated private Doctor Packs and a user-approved offline emergency card from verified information.
- implemented PostgreSQL row-level security, private object storage, audited RPCs, short-lived access paths, and SQLCipher-backed offline storage.
- localized core caregiver flows in English, Hindi, and Telugu without machine-translating original medical evidence.
- added strict TypeScript contracts, application tests, database linting, pgTAP security tests, and release validation.