Ujan Sen

WORK HISTORY

EDUCATION

ABOUT UJAN SEN

Hello there!I design and deploy high-performance AI systems that deliver measurable results — from real-time video super-resolution models running at 60 FPS on consumer GPUs, to GPT-powered tools that accelerate decision-making for millions of records, to multimodal Transformers improving clinical prediction.Key Highlights- Video Super-Resolution (AMD): CNN-based model for heavily compressed (QP-35) videos, outperforming internal and competitor baselines by +6% PSNR,+7% VMAF-7% LPIPS- Clinical Time-Series Prediction (Vector): Self-supervised masked Transformer trained on 150K+ MIMIC-IV patient records; AUROC 0.89, AUPRC 0.84-4% MAE; boosted by BioClinical BERT fusion- Supply Chain Optimization (Kinaxis): Graph ML pipeline with GPT-4 natural language interface; increased demand satisfaction by 400K+ units, reduced query time by 25%- Medical AI Research (Carleton): TCN model for surgical skill classification with 99% accuracy on real-world data; LLM suturing tutor improved accuracy by 85% in trials.Technical Skills: PyTorch, TensorFlow, CUDA, NetworkX, Python, Groovy, Bash, SQL, Transformers, Diffusion Models, Self-Supervised Learning, Graph ML, LLM Integration, Time-Series Modeling.Always happy to connect about computer vision, generative AI, applied ML in healthcare, and high-impact AI deployment. You can also catch me in intense discussions about soccer, video games, or mostly anything nerdy!