Home

Hi — I’m Shreyas Kulkarni
I am on the Applied Research team at Capital One, where I work on efficient language model inference. Previously, I worked on controllable synthetic image generation at Apple, and on diffusion model distillation for few-shot personalization at Adobe Research.
I am also a graduate student on short extended indefinite leave from the Computer Science department at UMass Amherst. One day, I shall return — like Odysseus — and finish my PhD. At UMass, I worked on privacy and security in federated learning. Before that, I studied Physics and Mathematics at IIT Guwahati.
My research interests lie in efficient generative modeling and machine learning systems. My current work explores diffusion and flow-based models for speculative decoding. On the systems side, I work on model quantization and inference optimization, developing low-latency serving pipelines for agentic workloads.