MatruSetu

Antenatal care, organised: assistive pregnancy tracking and OPD digitisation for OB-GYN clinics in India, in English, Hindi and Marathi.

Comment

The problem

A private OB-GYN practice in India can follow hundreds of pregnancies on paper registers. One missed anomaly scan or skipped follow-up can mean a missed complication. The bottleneck is continuity of care: noticing which mother is drifting out of follow-up and acting on it in time.

What it does

MatruSetu is an assistive pregnancy-tracking and OPD-digitisation app for OB-GYN clinics, in English, Hindi and Marathi.

  • Notice: enter the LMP once and the FOGSI/MoHFW-based ANC schedule (scans, tests, injections), gestational age and EDD build themselves; every mother is marked on track, at risk or lost from overdue items and contact history, with clinical alert filters.
  • Measure: visit recording with clinical flags (BP, Hb, FHR, fundal height), trend charts and a one-page printable report; photographed case papers and lab reports become a searchable timeline through in-browser OCR.
  • Act: a call queue with one-tap call or WhatsApp reminders in the mother's language. With the optional Gnani AI voice layer, staff can dictate vitals in Hindi, Marathi, English or Hinglish (a Devanagari-aware parser fills the form and the clinician confirms before saving), mothers receive spoken counselling, and WhatsApp voice notes pass through a danger-sign screen that replies urgently (nearest hospital / 108) and flags staff.

Method

Next.js 16, React 19, TypeScript and Tailwind CSS v4; Supabase (Postgres, Auth, Storage) with row-level security on every table; an installable PWA with an idempotent offline write queue; Tesseract.js and pdfjs-dist for OCR; Gnani AI Prisma (speech-to-text) and Timbre (text-to-speech), with browser speech APIs as a fallback. Risk status is rule-based and transparent (overdue items plus missed contacts), so a clinician can see exactly why a mother is flagged.

Results

  • Runs today: live demo at https://matruseva.vercel.app with a fabricated clinic of 10 synthetic patients (no real patient data).
  • 590 unit tests across 33 files covering clinical logic, scheduling, risk, i18n key parity, the offline queue, the WhatsApp bot and the voice client; CI runs lint, typecheck, SQL validation and a dependency audit (0 high vulnerabilities).
  • Accessibility: AA colour contrast, 44 px touch targets, keyboard and screen-reader support.

Evaluation

  • What we measured: correctness of the clinical logic, not patient outcomes. 590 automated unit tests (33 files) check that the ANC schedule, gestational age, EDD and follow-up risk status are computed as specified from FOGSI/MoHFW timings, including LMP corrections and overdue/missed-contact rules. All pass in CI.
  • Clinical metric (not yet measured): the outcome MatruSetu targets is the share of due ANC contacts completed on time (on-time ANC completion rate, %), and the number of mothers moved from "lost" back to "on track" per month. We have not run a clinic pilot, so we report no value for either; a before/after pilot in one clinic is the next step.
  • Cost: runs on free tiers (Vercel + Supabase) for a single clinic; the optional Gnani voice layer uses free API keys within their limits, with browser speech APIs as a zero-cost fallback.

Who benefits, and who could be harmed

  • Benefits: pregnant women in private OB-GYN clinics who might otherwise miss scans or follow-ups; clinic staff who currently track hundreds of pregnancies on paper; mothers more comfortable in Hindi or Marathi than English.
  • Could be harmed: a mother could be harmed if staff trust a wrong schedule or a missed flag instead of their own judgement, so every flag is labelled assistive until an OB-GYN signs off the protocol. Untranslated or poorly translated counselling could mislead. Health data (including voice sent to Gnani) is sensitive: we use row-level security, signed expiring links and fail-closed configuration, and the mother's consent is required before voice features are used.

Where it fails (limitations)

  • Assistive, not diagnostic. The ANC schedule and risk thresholds follow FOGSI/MoHFW guidance but are awaiting sign-off by a practising OB-GYN; until then every flag is presented as a prompt, not a diagnosis.
  • The Hindi and Marathi wording has not yet been reviewed by native-speaker clinicians.
  • Voice accuracy and latency depend on the Gnani key and audio quality and have not been formally measured.
  • The demo uses synthetic data only; there is no clinical outcome evaluation yet.
  • Proactive outbound WhatsApp reminders need Meta Business verification; today the bot answers inbound messages only.

Team: Madhav Zanwar, Divyani Papalkar (MatruSetu was first built for the IIT Bombay Health-a-thon 2026 with Veerbhadra Mahant and Hari Chaudhari). Code (public): https://github.com/madhavzanwar/Matruseva

Built With

Share this project:

Updates

Submission history