Hamid Reza Hassanzadeh

• Leading two key Generative AI projects focused on processing medical unstructured texts at enterprise scale • Architecting and deploying production-grade LLM pipelines for healthcare applications serving millions of members • Designed and implemented RAG-based systems for medical document understanding, reducing manual review time significantly • Driving research on agentic AI workflows for clinical decision support, resulting in multiple patent filings • Fine-tuning and evaluating open-source LLMs (LLaMA, Mistral) on domain-specific healthcare tasks using distributed training (FSDP, • DeepSpeed) • Filed 3+ patents on mixture-of-experts language modeling, synthetic QA generation, and question resolution techniques • Collaborating cross-functionally with clinical SMEs, engineering teams, and leadership to translate research into deployed products

• Conducting independent research on prediction of suicide events from unstructured clinical texts using large language models • Pioneered in-context learning approaches for suicide event prediction, achieving state-of-the-art performance on clinical benchmarks • Co-authoring multiple publications on LLM-based clinical NLP with collaborators across institutions • Mentoring graduate students on applied NLP research in healthcare settings

(Formerly 3M Health Information Systems / MModal) • Designed and implemented transformer-based NLP systems for generating SOAP notes from patient-doctor conversation transcripts • Built a concept-induced AI system that reduced hallucination rates in generated clinical summaries by grounding outputs in verified medical concepts • Improved summary quality by integrating explainability techniques (XAI), enabling clinicians to trust and validate AI-generated notes • Developed efficient context fusion methods for generating holistic summaries from full-length clinical transcripts • Published research at EMNLP Findings and Interspeech on leveraging pretrained models for doctor-patient conversation summarization

• Built an end-to-end NLP pipeline for extracting actionable insights from survey data generated by streaming service subscribers • Developed deep learning-based sentiment analysis models to quantify subscriber satisfaction across content categories • Implemented transformer-based extractive summarization to condense thousands of open-ended survey responses into executive-ready reports

• Developed DeepClick, a deep NLP ranking model that improved search engine result relevance by leveraging user interaction signals, item descriptions, and search terms • Engineered the full pipeline on Hadoop/Spark infrastructure, integrating RNN-based architectures with word embeddings and Siamese networks • Processed and analyzed large-scale clickstream data to train and evaluate ranking models at production scale

• Designed and built a deep learning-based recommender system for suggesting shopping destinations to drivers based on driving behavior and personal preferences • Created a novel end-to-end trainable architecture fusing collaborative filtering and content-based recommendation approaches • Deployed on Hadoop/Spark with TensorFlow, processing real-time geolocation and behavioral data