Anuj Iravane is Head of AI at Anterior, building clinical-reasoning systems for healthcare decisions that require accuracy, explainability, and human oversight. His central challenge is evaluating automated medical decisions when sensitive patient records cannot be retained.
After studying computer science at Northwestern University, Iravane worked on recommender systems at Amazon before moving into healthcare AI. At Anterior, he develops systems for prior authorization, payment integrity, and quality measurement, where fragmented clinical documentation must be interpreted against complex policies and unusual patient histories. His public recruiting for AI researchers and engineers emphasizes clinical accuracy, explainability, and dependable production systems.
Making clinical AI accountable
- Guideline Decision Trees. Iravane’s research on structured guideline execution translates medical policies into explicit conditions, dependencies, and outcomes. Separating policy logic from clinical-evidence interpretation makes consequential decisions easier to inspect and supports accountable human review.
- Synthetic clinical data with predetermined outcomes. When privacy restrictions prevent retaining real records, Iravane starts with a desired decision, samples reasoning paths through a symbolic policy tree, and generates patient histories consistent with those conditions. This reverse-conditioned evaluation approach exposes rare clinical scenarios that small customer datasets may miss.
- Coarse-to-fine medical-record generation. His pipeline establishes patient characteristics, constructs clinical timelines, plans provider encounters, and produces individual documents. Automated checks detect contradictions, while round-trip evaluations assess whether generated records support their intended decisions. Clinicians steer generation and adapt modular workflow components without requiring application-code changes.
- Error-rate-based fairness. In research coauthored with Sai P. Selvaraj and Khadija Mahmoud, Iravane evaluates fairness against human-reviewed clinical decisions instead of assuming demographic groups should have identical approval rates. Their analysis covers 7,166 cases across 27 guidelines; findings concerning race and ethnicity remain inconclusive because subgroup data was limited.
His work makes healthcare systems testable before deployment while keeping clinical experts responsible for the scenarios, judgments, and limitations that matter most.
