Explicit safety gates
Less apparent autonomy in exchange for clearer escalation.
Materna
A safety-oriented agentic system for structuring obstetric consultations and making escalation paths explicit.
- Python
- FastAPI
- Multi-agent orchestration
- Clinical safety
Where should
automation
stop?
An obstetric consultation is a setting where missing information matters. Materna explores how agents can structure information, make handoffs inspectable, and keep review and escalation explicit.
Materna · research workflow
Care
before
clever.
before
clever.
StructureReviewEscalate
The review board
The working dashboard. Smart Paste on the left, the four-agent pipeline on the right — Intake, Dating & Risk, Guideline, Auditor — and the Band room where every handoff is posted. Every case shown is synthetic.
Local build on synthetic demo cases, September 2026
Handwriting to structure
The intake step in the shipped product: clinical notes or a photograph of a handwritten chart, in English or Urdu, normalised into a typed case before any reasoning happens.
Production build, captured September 2026
The disclaimers, in the product
The limits are stated inside the product, not only in the write-up: synthetic data only, decision-support rather than diagnosis, and an explicit warning that it is not HIPAA certified and must not be used with real patient data.
Production build, captured September 2026One case, four agents, one human
How a case reaches a decision, and who is allowed to make it.
Free text, a form, or a photograph of a handwritten chart in English or Urdu becomes a validated, typed case. Nothing downstream has to guess what a field meant.
Decisions.
And their cost.
Structured artifacts between agents
More implementation discipline in exchange for inspectable handoffs.
Voice-to-note assistance
A narrower product surface in exchange for practical workflow value.
What's
in the system.
- Structured clinical workflow exploration
- Safety-first multi-agent orchestration
- Designed around reviewable rather than hidden decisions
Materna is clinical AI research. Its public story is about workflow design and evaluation, with clinical governance still essential to any deployment.
What this
proves. And
what it doesn't.
- Verified in sourceMulti-agent orchestration with structured, inspectable handoffsVerified in sourceMaterna repositoryFastAPI service with multi-agent orchestration; structured handoffs are architectural, not cosmetic.
- Verified in sourceSynthetic data only, stated inside the productVerified in sourceMaterna safety section, in the shipped productThe build states its own limits on screen: synthetic data only, decision-support rather than diagnosis, and an explicit warning that it is not HIPAA certified and must not be used with real patient data.
- Not yet provenClinical readinessNot yet provenResearch stageThis is workflow research on synthetic cases. Clinical validation, clinical governance, and deployment for patient care have not been established by this portfolio audit. A public hackathon demo is not clinical deployment.
The next
hard question.
The work is intentionally research-oriented; evaluation and clinical governance must keep leading product ambition.
Into the research