The Role

5C's AI is judged against what its radiologists say is correct. This role owns that standard. You will design the gold-standard benchmark datasets, adjudication protocols, and ground-truth labeling standards that Bionic's models are measured against — turning clinical judgment into evaluation the Applied AI team can build to.
You are an NMC-registered radiologist with rigorous reporting standards and an interest in how models fail. Any subspecialty is welcome; depth in MSK, neuroradiology, or chest imaging is particularly valued. You will not be working through a reading worklist — you will define what a correct read looks like, case by case, so model performance is measured against something real.
The pitch is simple: your expertise, amplified. A single well-adjudicated benchmark case shapes the behaviour of models that read millions of scans. This is the radiologist who teaches machines what excellent looks like.
We are strongly biased toward candidates who can join quickly. If your notice period is short, your application will move faster because these teams are being built now.

What You Will Own

Gold-standard benchmark datasets: case selection, cohort design, difficulty stratification, and subspecialty coverage across CT, MRI, X-ray, and mammography
Adjudication protocols: consensus panels, tie-break rules, escalation paths, and documentation that makes every ground-truth decision auditable
Ground-truth labeling standards: finding definitions, severity scales, laterality and measurement conventions, and rubrics precise enough that a second radiologist applying them reaches the same answer
Inter-reader variability: measuring it, deciding when disagreement is signal versus noise, and designing consensus processes that converge without flattening legitimate clinical judgment
Model failure review: working daily with the Applied AI team to examine where models diverge from ground truth, and deciding whether the model, the label, or the benchmark is wrong
Evaluation criteria: translating clinical judgment into measurable definitions of correctness — what counts as a miss, what counts as a clinically irrelevant difference, what counts as false reassurance

MD or DNB in Radiology with active NMC registration
Rigorous personal reporting standards and the ability to articulate why a read is correct, not just what the read is
Comfort with structured criteria: you can turn "I know it when I see it" into a written standard another radiologist can apply consistently
Interest in how AI models behave and fail. You do not need to write code, but you need to reason clearly about sensitivity, specificity, false positives, false negatives, and calibration
The temperament for adjudication: patient with disagreement, precise in documentation, and comfortable being the final word on ground truth

Subspecialty depth in MSK, neuroradiology, or chest imaging
Experience with reader studies, research methodology, or statistics: kappa, ROC analysis, cohort design
Teaching or peer-review experience — you have already spent time explaining what separates an adequate read from an excellent one
Prior exposure to AI model evaluation, annotation programs, or clinical quality assurance

Your judgment scales. A benchmark you adjudicate once shapes model behaviour across every scan those models read
You sit at the centre of the Applied AI team, not at its edge. Ground truth is the constraint everything else is built around
5C reads scans for 2,000+ hospitals and diagnostic centers, so the standards you set propagate through one of the largest radiology operations in the country
We are strongly biased toward candidates who can join quickly. If your notice period is short, your application will move faster because these teams are being built now.

Apply for this role