Razi Mahmood

Experience

Experience

Applied multimodal AI, medical imaging research, text analytics, and ML infrastructure.

Recent HOPPR

Machine Learning / AI Research Intern

  • Developed a head CT radiology report generation model using fine-grained finding prediction and probabilistic patient-similarity retrieval.
  • Built a non-LLM generation approach designed to improve clinical grounding and reduce hallucinations.
  • Demonstrated improved accuracy and lower hallucination rates compared with LLM-based report generation baselines.
  • Worked across multimodal medical AI, computer vision, clinical retrieval, radiology text generation, and model evaluation.
May 2023 - June 2025 RPI Deep Image Analysis Lab

Graduate Researcher

  • Developed discriminative fact-checking models to verify whether AI-generated chest X-ray reports are clinically consistent with medical images.
  • Built image-driven and phrase-grounded methods for evaluating and correcting generated radiology reports.
  • Created RadCheck, an open-source dataset of real/fake image-finding pairs for report verification.
  • Demonstrated 15-40% improvements in AI report quality across experimental settings.
  • Ran large-scale model training and evaluation workflows on Slurm-managed GPU clusters, including distributed experiments for medical image-text modeling.
  • Published work across MICCAI, ISBI, MLMI, NeurIPS GenAI4Health, and CVPR; one patent filed.
July 2025 - October 2025 IBM Storage Systems

Data Science Intern

  • Built search evaluation tooling for a content-aware RAG storage product.
  • Integrated search evaluation benchmarks into daily regression and CI/CD testing workflows used by developers, performance testers, and QA automation teams.
  • Developed LexLogit, a re-ranking model combining semantic, lexical, and re-ranker signals.
  • Worked with vector databases, search benchmarking, MLOps, and Python automation.
Sept. 2021 - April 2022; earlier internships 2014 - 2016 IBM Almaden Research

Machine Learning Research Intern

  • Co-developed a computer vision method for detecting localized abnormalities in chest radiographs.
  • Improved anomaly detection AUC by 10% on MIMIC chest X-ray data.
  • Published at IEEE ISBI and contributed to patent filings.
  • Contributed to statistical and deep learning methods for cardiac imaging analysis.
Jan. 2021 - May 2021 IBM Watson

Machine Learning Research Intern

  • Developed a log anomaly classification model using BERT and supervised contrastive learning.
  • Collaborated with IBM Watson AIOps researchers on text analytics for IT operations.
  • Published work at IEEE BigData-IT.
Aug. 2021 - Nov. 2021 Hyperfine

Data Science Intern

  • Built NLP-based data annotation workflows for radiology reports.
  • Extracted 7,200 annotations from 600 brain MRI reports using sentence parsing and phrasal grouping.
  • Created labeling tools that reduced annotation time by 10x.
June 2021 - Aug. 2021 Xoran Technologies

Data Science Intern

  • Developed 3D medical image segmentation models for cone-beam CT.
  • Trained U-Net models for head and neck anatomical segmentation.
  • Supported clinical data labeling workflows using ITK-SNAP and 3D Slicer.