Akforges
← All work
Mobile— Women's Health · India
FemmeHealth
Digital Therapeutics · India

PCOS. Tracked, correlated, explained by AI.

A comprehensive Android digital therapeutic for women with PCOS — eight health modules, a 30-day Pearson correlation dashboard, AI health insights from Claude, and a separate web portal for doctors. All offline-first.

8
Health modules
80+
Indian foods in DB
Claude Haiku
AI engine
Offline-first
Architecture

The problem

PCOS affects roughly 1 in 5 women in India — one of the highest prevalence rates globally. Yet most women manage it with fragmented tools: a period tracker here, a nutrition log there, a notes app for symptoms, and whatever their gynaecologist remembers from the last appointment.

The deeper problem is that PCOS is a complex, multi-system condition. Sleep quality affects symptom severity. Workout intensity interacts with cycle phase. Calorie intake correlates with lab markers. No single tracking app connected these dots — they just logged data without insight.

The brief was to build a medically-informed digital therapeutic that could surface these correlations automatically and explain them in plain language — bridging the gap between patient data and the gynaecologist's appointment.

What we built

Eight distinct health modules, each built with offline-first Room database storage and background Supabase sync: PCOS screening (4-section questionnaire with risk scoring), daily symptom logger (5-metric scoring with 30-day trend graphs), cycle tracker with Ogino ovulation prediction, lab investigation logging with auto-interpretation of 6 key markers (LH/FSH ratio, HOMA-IR, HbA1c, androgens, thyroid, lipid panel), nutrition logging with 80+ Indian foods with ICMR macros, medication adherence tracker, sleep tracker with Health Connect sync, and cycle-phase-aware workout recommendations.

The correlation dashboard is the centrepiece. It computes Pearson correlation coefficients across 30-day windows — sleep quality vs symptom severity, workout intensity vs mood scores, calorie intake vs cycle regularity, and more. Any |r| ≥ 0.3 surfaces as an actionable insight. Claude Haiku analyses six defined correlation rules and generates plain-language explanations contextualised to the user's specific data.

Smart alerts run via WorkManager daily jobs — spike/deficit/streak/overdue notifications that are delivered even when the app is backgrounded. PDF health reports export the full 30-day data set with AI-generated insights and a medical disclaimer, formatted for sharing with a gynaecologist.

The doctor portal is a separate React 18 web app with a consent-based patient data access workflow — doctors see the correlation charts and timeline, not raw personal details, until the patient explicitly grants access.

Architecture

Android app uses MVVM + Clean Architecture with Hilt for dependency injection. Room is the source of truth — all reads go to Room, a background SyncWorker replays queued changes to Supabase when connectivity returns. Supabase enforces row-level security so no user can access another's data even with a valid JWT.

The lab interpretation engine (InvestigationInterpreter) applies clinical reference ranges to flag abnormal values — it was built with the assumption that a gynaecologist reviews findings before any action is taken, and the UI makes this explicit with appropriate disclaimers.

Tech stack

Android (Kotlin)Jetpack ComposeMaterial 3HiltRoomSupabaseClaude Haiku (Anthropic)Vico ChartsReact 18TypeScriptViteRechartsFirebase AnalyticsHealth Connect
Build something similarNext case →