01
Offline AI inference
Seven TensorFlow.js models run in-browser with no server round-trip, with FastAPI fallback for low-powered devices.
All work
Healthcare Screening PWA · 2025
NCD Health AI
A clinical screening tool for detecting non-communicable diseases, deployed in South Africa.
Highlights
7 AI screening models
Offline submission queue
District-scoped admin analytics
Case study
What made this build matter.
02
Offline submission queue
Submissions captured offline are stored in IndexedDB and delivered automatically through Background Sync.
03
Push notifications
Workers opt in to VAPID push while admins target messages by district or risk level and track opens.
04
Follow-up reminders
BullMQ schedules stable 24-hour reminder jobs and daily sweep jobs for weekly and monthly follow-ups.
05
Jurisdiction dashboard
Regional admins see risk data, breakdowns, and heatmaps strictly scoped at the database query level.
06
Feedback loop
Patients report follow-up and assessment accuracy, feeding report quality metrics visible to admins.
07
PWA service worker
Workbox handles cache-first static assets, network-first requests with offline HTML fallback, and reliable cache invalidation through service worker activation broadcasts.
Interface views
A closer look at the product surface.
Interface slide 1 / 2
NCD Health AI screening and vitals capture interface
Product surface
The important screens are not decoration. They show the operational surface: what users do, what admins control, and where the product has to stay reliable.
7 AI screening modelsOffline submission queueDistrict-scoped admin analytics
The problem
Healthcare workers needed to collect patient vitals and symptoms, run AI inference, generate structured health reports, and serve regions where connectivity can be unreliable or unavailable.
What we built
We built an offline-capable PWA with browser-based AI inference, IndexedDB submission queues, Background Sync delivery, VAPID push notifications, BullMQ reminder jobs, and jurisdiction-scoped admin analytics.
The outcome
Healthcare workers can screen patients without connectivity, while public health administrators monitor aggregate risk data through regional, district-scoped dashboards.
Technical note
Seven condition-specific TensorFlow.js models run entirely in the browser for offline inference, with FastAPI server-side fallback for low-powered devices.
Stack
ReactViteTypeScriptExpress.jsFastAPIDrizzle ORMPostgreSQLTensorFlow.jsBullMQRedisSupabase RealtimeWorkboxVAPID PushRailwayVercel
Next project
Theranope
Start your project
Want a build like NCD Health AI?
Share the workflow, product idea, or operational problem you want solved. We'll reply with a clear next step.
Book a meetingBuild something similar
Want a build like NCD Health AI?
Tell us what part of this case study matches your own product or workflow.
What are you looking for?