Vaidya AI
Fractal Analytics needed a production-grade consumer health app — not a prototype. Vaidya AI unifies medical records, HealthKit/Health Connect fitness data, AI-powered health insights, and subscription billing into a single platform. Presented at the India AI Summit. I owned the mobile architecture end-to-end and built the FastAPI payment backend.
Owned the mobile team's base project template — folder structure, naming conventions, Husky pre-commit hooks (lint/format/type checks) — later adopted across all subsequent Flutter projects at Ailoitte.
Reduced cold start from 3.5s to under 1s (71% improvement) and binary size by 35% — via parallelised initialisation, trimmed dependency tree, and eliminating a main-thread serialisation bottleneck identified through Firebase Performance Monitoring.
Designed dual-platform health data sync: HealthKit (iOS) triggered by HKObserverQuery + enableBackgroundDelivery; Android triggered by WorkManager PeriodicWorkRequest. One shared SyncEngine in Dart. Idempotent upserts keyed on deterministic client_record_id — retries and duplicates are safe by design. 10,600+ entries synced per user.
Built end-to-end mobile observability: Crashlytics for both fatal and non-fatal errors, with Dio log breadcrumbs wrapping every API/asset failure, sliced by app version, OS, and device. Triaged 180+ production events. Used this to trace a zero-repro iOS launch white screen (malformed token + corrupted navigation state) to root cause.
Built FastAPI + PostgreSQL payment microservice: idempotent Razorpay webhook handler with HMAC-SHA256 verification, order creation, automated refunds, subscription state machines for recurring billing, atomic credit deduction (WHERE balance >= cost in a single SQL statement).
Security hardened against SOC audit: SSL pinning, root/jailbreak detection, code obfuscation, OAuth injection fix, HTTPS App Links + deep-link hosting. Root-caused an iOS Keychain persistence bug (credentials surviving uninstall) via fresh-install detection under accessibility mode analysis.
Debugged production Apple SSO userId=0 bug — one user, no repro, super_admin data leaking. Traced via log breadcrumbs to stale SQL query missing isactive=1 after soft-delete shipped. Fixed. Immediately hit a second bug: Apple SSO returns the same token tid every time, triggering a Redis login-block that Google SSO never hit. Both resolved in 24 hours.
Presented at the India AI Summit. Live on App Store (5K+ downloads, ranked under top 110 medical apps on iOS) and Play Store (20K+ downloads). Cold start improved 71%. Binary size reduced 35%. Zero payment-related incidents since FastAPI microservice shipped.