Medtech developer
under NDA
Industry
healthcare · an early cancer-diagnosis product
Duration
22 wks
Team
not published
Disciplines
AI · Security · Infrastructure · Development
Task
Get the first version of the algorithm into daily use in a hospital: embed it in the HIS, clear the secured perimeter, retrain the models.
Result
1.4% of false-negative reports sent back for review thanks to the algorithm. A pilot in a working hospital: the tool runs inside the HIS, the clinician sees the result where they already work.
How it went
022 wks
stage 1 · 14 wks
A product around the algorithm, integrated with the HIS
DevelopmentSecurityInfrastructure
Task
The first version of the algorithm had no product around it: no clinician interface, no path into the HIS, no clearance for the perimeter. We worked through the medical formats — CT, MRI, X-ray — and the internals of several HIS platforms. We built patient-data handling to meet the secured perimeter’s requirements.
Result
The algorithm runs inside the HIS and is cleared into the hospital’s secured perimeter. The clinician sees the result where they already work.
stage 2 · 8 wks
Models, MLOps and a team in the hospital
AIDevelopment
Task
Refined the algorithm and retrained the models on the hospital’s data. Set up the MLOps pipeline: data → training → rollout → quality checks. Built the team and the forward-deployed-engineer processes — engineers working next to clinicians.
Result
Model updates ship through the pipeline, not by hand. The hospital’s team was hired through our search — some of the engineers work on site: supporting the tool and studying how clinicians use it.
A similar problem?
In 45 minutes we'll find where your bottleneck is and what we'd do about it in two weeks. Free, with no presentation about us.