AI for medtech, pharma and medical research
Custom models for biomarker discovery and medical research, built for validation from the first commit and documented for the approval path.
Email us
The problem with AI in regulated medical work
Evidence, not demos
A model without an agreed validation scheme is evidence of nothing. In medical work that is not a nuance, it is the whole question.
Data you cannot move
Provenance, consent and residency constrain the architecture before anyone picks a framework. We design around them rather than against them.
Documentation debt
Approval documentation written after the fact gets written twice, and the second version rarely matches the code.
Custom ML for biomarker discovery
Image-based models
Models trained on stained tissue and other imaging modalities, including whole-slide pathology.
Validation first
Splits, metrics and reporting are agreed before training starts, and the split is held out until the end.
Data, ethics and approvals
We work inside the governance you already have and write the documentation as the work happens, shaped for the approval path the project sits in.
Two case studies
Treatment response from biopsy scans
An AI and ML pipeline that trains a model on H&E-stained cancer biopsy scans to predict a patient’s chance of responding to treatment.
Health real-world-data joint venture
Setup and concept for a joint venture in health real-world-data research: structure, scope and operating model.
Questions
Do you work under an approval regime?+−
We write the documentation alongside the code so it fits the approval path the work sits in. Tell us which regime applies and we will say plainly what we have done under it and what we have not.
Can our data stay where it is?+−
Yes. We design the pipeline around the data residency and governance you already have, rather than asking you to move anything.
What do you need to start?+−
A described task, a clear definition of the label, and access to a sample. That is enough for us to say whether the problem is tractable.
Who owns the model?+−
You do. You get the code, the weights and the training record, and nothing in the pipeline depends on us staying involved.
Still have questions? Get in touch