Brief
Built for my father, who manages diabetes, BP, and daily medications with limited digital comfort and a preference for speaking Hindi over typing. Every design decision flows from one question asked before writing a single line of code: does he actually log daily? Phase 1 answers that question — frictionless glucose logging via voice or numpad, a streak-based habit loop, medicine reminders, and a safety bypass that cannot be silenced when readings cross critical thresholds. The app understands how he actually speaks — colloquial Hindi, not textbook input. It works fully offline and syncs when connectivity returns. Family members get a read-only monitoring layer — aware without being intrusive. Subsequent phases layer in cross-condition insights, AI-powered health summaries, and deeper guardian alerting only after the retention habit is proven. Built as a production-grade system with safety-critical paths at 100% branch coverage and domain logic tested exhaustively as pure functions — no database, no mocks, millisecond feedback.
Challenge
- No existing health app is designed for elderly Indian users managing multiple chronic conditions simultaneously
- Voice input must understand how people actually speak Hindi — colloquial, regional, imprecise — not standard speech output
- Safety-critical alert paths must execute completely even when individual delivery channels fail
- Habit formation must be proven before intelligence is built — premature AI with no retention is wasted engineering
Approach
- Phased architecture: retention first, intelligence second — each phase gated by a measurable human behaviour metric
- Voice parser built as a pure function fully decoupled from device hardware — exhaustively tested before any mic integration
- Offline-first mobile sync so unreliable connectivity is never a reason to miss a log
- Domain logic (streak engine, feedback engine, voice parser, notification resolver) isolated as pure TypeScript with zero database dependencies — tested at millisecond speed
- Multi-channel safety alert system designed for resilience under partial failure
Outcome
- Phase 1 shipped: glucose logging, voice pipeline, habit loop, streak engine, medicine reminders
- 100% branch coverage on all safety-critical paths
- Pure domain logic tested exhaustively — hundreds of edge cases, sub-second test suite
- Voice pipeline validated for colloquial Hindi input patterns before hardware integration