RadCheck
Large-scale open-source dataset for radiology report verification using image-grounded real/fake finding-location pairs.
View on Hugging FaceBiomedical Engineering Ph.D. Student at UBC
Razi Mahmood
Trustworthy multimodal AI researcher working across medical imaging, computer vision, NLP, model evaluation, retrieval, and data annotation tooling.
Selected Work
Research depth with implementation range.
Radiology report verification
Discriminative fact-checking models that verify whether generated chest X-ray reports are clinically consistent with medical images.
See related papersHead CT generation at HOPPR
Non-LLM report generation using fine-grained finding prediction and probabilistic patient-similarity retrieval to improve accuracy and reduce hallucinations.
Read experienceSearch and RAG evaluation
CI/CD-ready evaluation workflows for retrieval quality, vector search, re-ranking, and regression testing in applied ML systems.
Read experienceAbout
Current direction
I am a first-year Ph.D. student in Biomedical Engineering at the University of British Columbia, advised by Prof. Purang Abulmouselmi. My interests include multimodal AI, computer vision, medical imaging, natural language processing, and ML systems for reliable model evaluation.
My recent M.S. thesis research at RPI focused on trustworthy AI for radiology, especially discriminative fact-checking models that verify whether AI-generated chest X-ray reports are clinically consistent with medical images. This work led to multiple peer-reviewed publications, a filed patent, the RadCheck dataset, and the 2026 UC Berkeley Data Science Changemaker Award.
On the implementation side, I work with Python, PyTorch/TensorFlow, Slurm-managed GPU clusters, distributed training and evaluation workflows, CI/CD testing, Git/GitHub, VS Code, Jupyter, and Claude-assisted development.
MICCAI, ISBI, CVPR, NeurIPS
Recent venues and workshops
28M+
RadCheck rows on Hugging Face
15-40%
Reported quality improvement in AI reports
2026
UC Berkeley Data Science Changemaker Award