Proof of concept · Shanti Infosoft

Know what moves your numbers.

Everyone with high blood pressure gets the same advice: sleep more, eat less salt, stress less. Nobody gets told which one is actually raising their reading — or by how much.

Open the member app Explore all 64 screens
Built to the 2025 AHA/ACC guideline POPIA & GDPR by design Model fits live in your browser
—
of one member’s blood-pressure variation explained by their own habits
—
days of paired readings and check-ins fitted
45s
daily check-in budget — past that, people stop logging
1 of 5
factors withheld, because its interval still crosses zero
The gap

Population advice, personal problem.

The 2025 AHA/ACC guideline makes lifestyle change first-line treatment for every adult with raised blood pressure, and for the first time names stress management alongside diet, activity, sodium and alcohol. But a guideline speaks to populations. It cannot tell one person that their systolic climbs 9 mmHg the morning after a short night, while their salt intake barely registers.

Closing that gap — between advice that is true on average and advice that is true for you — is the whole product.

The daily loop

Sixty seconds a day. That’s the whole ask.

Every design decision is subordinate to that budget, because the moment the loop takes longer the data stops arriving and the model has nothing to learn from.

1 Prompt

A push at 06:30Quiet hours enforced server-side, so a wrong device clock can’t cause a 3am buzz.

2 Capture

Classified as you typeAgainst the real thresholds — and above 180/120 it stops being a save flow.

3 Understand

Ranked, quantified, honestEvery figure carries a confidence interval and the sample size behind it.

The part that either works or it doesn’t

A number, an interval, and the nerve to say nothing.

A per-user model fits each person’s readings against their own lifestyle features, with every factor lagged to when it should actually act — alcohol overnight, sleep debt accumulating, sodium the following morning.

The output is a coefficient in mmHg. That is why the app can say “about 9 mmHg” instead of “sleep may affect your blood pressure”.

Movement is deliberately missing from this chart
Its interval runs from −1.9 to +0.1 — it crosses zero, so the model is not allowed to claim it. Roughly a third of candidate insights never reach a member. That restraint is what makes the other four worth reading.
See it in the app How the model works
Effect on next-morning systolic 95% intervals

Fitted in your browser at page load from 90 days of readings. — of the variation in this member’s morning systolic is explained by these five factors.

Beyond the loop

A whole platform, not one clever screen.

Everything below is built and clickable in this prototype.

Trends & patterns Ninety days against the clinical bands, morning-versus-evening split, variability, and every reading plotted on the category grid. Open trends → What-if simulator Drag sleep down to five hours and watch the projection move by exactly the coefficient the model measured. One model, run in both directions. Try the sliders → Single-subject experiments Change one thing for three weeks and get a pre/post readout. Proof that a recommendation worked, for this person. See experiments → Reports a GP can use A monthly summary that exports as FHIR R4, so a practice system ingests it without anyone re-typing numbers off a PDF. View the report → A crisis path with no AI in it Above 180/120 the model is bypassed entirely — deterministic threshold, immediate instruction, and an operational alert with a 30-minute SLA. See the safety path → Operations console Eleven pages: engagement, cohort analytics, the clinical safety queue, model governance, and the compliance tooling POPIA actually requires. Open the console →
Your data never leaves South Africa
POPIA §72 has no adequacy list, so instead of building a transfer regime we host in AWS Cape Town and remove the question.
Consent per purpose, revocable
Health data is special personal information. Six separate decisions, a visible ledger, and a working export and delete path.
A wellness product, not a medical device
It records, analyses and educates. It does not diagnose, does not treat, and never advises on medication. The roadmap names exactly where that line sits.

All of it is clickable.

Sixty-four screens, an operations console, and the full proposal — stack, modelling approach, compliance, timeline, team and a module-by-module cost estimate.

Explore the prototype See the estimate

Or press ⌘K on any screen to see which line of the estimate it belongs to.