October is Breast Cancer Awareness Month. Learn the signs and take the pledge

Our work · Project 01

In development

Health tools that assume nothing.

No connection, no reading ability, no clinic down the road. The EquiAccess Indexer is a private, offline health tracker and triage guide, built first for rural sub-Saharan Africa and South America.

Try the demo How a visit works
Two health workers reviewing information on a tablet
3.16Bpeople can't reach a health facility within an hour on foot.
646Mstill can't, even with motorised transport.

So the first steps of care have to come closer to home. See the map

What the Indexer does

Tracks health over time

Remembers every visit in a private, encrypted record on the device, so patterns show up that one visit can't.

AI-assisted, never AI-decided

A fine-tuned Llama 3.2 helps word the questions. Clinical pathways and danger signs are fixed rules written by people. Built with Llama.

Private, offline, inclusive

Runs with no internet. Speech in and speech out, so it works at any literacy level or dialect.

How a visit flows

  1. 1Verify and take vitalsDate of birth plus a PIN or keyphrase, then basic vitals.
  2. 2Fixed symptom checkThe same questions every time: energy, mood, and physical state.
  3. 3Guided questionsA clinical pathway such as OLDCARTS, phrased with AI help.
  4. 4Danger-sign flagsFixed rules. The AI never decides them.
  5. 5EquiAccess IndexA transparent priority score for a clinician to review.
  6. 6ReportWeighed against past visits and saved on the device.

Deterministic engineAI helps with phrasing only

Live demo

One visit sees a moment. A month sees a pattern.

Any single week can look fine. The Indexer watches across weeks and compiles what one visit can't see, because it remembers the patient. Pick someone and step through their month.

—
EquiAccess IndexMonitoring
Week 1This visit alone

Week 1 of 4

Not a diagnosis. The Index is an information-gathering summary for a clinician; cross-visit flags and weights are drafts pending clinical review. Patients are scripted, but every flag and score is exactly what the engine produced.

What the AI learned from

No synthetic data, ever.

The model learns to phrase questions only from real clinical conversations between real people, used under licenses that permit it. What to ask, and which signs are dangerous, is written by people from public-domain MedlinePlus references, never trained into the model.

MIT

MTS-Dialog

Real doctor and patient clinical dialogues.

3,500+ real doctor questions
CC BY 4.0

PriMock57

57 primary-care consultations recorded with real clinicians.

1,900+ real doctor questions

Project 02

Upcoming

Digital literacy and accessible software

An easy-to-use platform for learning digital skills, health-related and beyond, and for getting assistive software into more hands.

Learn. Digital-literacy lessons for every starting point.
Access. Assistive software, such as screen readers, for the people who need it.
Equip. Later: assistive hardware such as wheelchairs and hearing aids.

Build it with us.

Clinician, translator, engineer, or community partner: we design with the people we design for.