The Evidence Layer: agentic AI in healthcare
Why agentic AI in healthcare needs an evidence layer that clinicians can inspect, question and trust.
Read the articleResearch/Scientific intelligence
We study how intelligent systems can be useful, practical and accountable in scientific and healthcare settings, where trust is part of the product.
Four commitments shape every programme, from the first experiment to deployment.
Outputs are anchored in sources a reviewer can open and check.
Systems show how they reached a conclusion, not just the conclusion.
Expert judgement stays central. AI supports it and never replaces it.
We evaluate against real institutional outcomes, not demo quality.
Long-running research areas that feed our products and engineering work.
Methods for working across documents, tables, voice, imagery and structured scientific information.
Interaction models that make review, correction and expert judgement central to the workflow.
Reliable ways to extract, contextualise and use rich scientific and operational documents.
Practical approaches to privacy, transparency, evaluation and institutional readiness.
Research at NMI is applied: every programme is expected to survive contact with real users.
A concrete institutional problem, defined with the people who live with it.
Small systems that test the riskiest assumption first.
Rubrics, expert review and outcome metrics, not anecdotes.
Only what holds up under evaluation becomes a product.
We partner with institutions on applied research that can become real systems.