Nirajan Kunwor

I am a Senior Software Engineer at PortPro, with research interests in machine learning robustness and medical AI — domain generalization, test-time adaptation, federated learning, and parameter-efficient fine-tuning for real-world clinical deployment.

My recent work studies how vision-language and foundation models fail to generalize across skin tones in dermatology AI, and how federated, parameter-efficient adaptation (LoRA) can align models like BiomedCLIP across international chest X-ray cohorts without centralizing patient data. I’m also working on test-time adaptation methods that resist prediction collapse under distribution shift.

Before this, I spent several years as a full-stack/backend engineer building production systems — identity verification workflows, EDI integrations, and large-scale data migrations — at PortPro, BeyondID, and Spiralogics.