Resume ⋙

Hello / namaste!

I am Pavan. I currently work on optimizing deep learning model inference and GPU kernels using agentic methods at AMD.

Highlights

GEAK • Autonomous multi-agent system that makes AMD GPUs run faster — it profiles the underlying kernels of a given model in a vLLM/SGLang serving stack, then generates, tunes, and validates optimized kernels across Triton, HIP, FlyDSL and other languages. [GEAK v4] [GEAK v3]

---

Prior to this, I worked at Amazon in improving LLM-based speech-to-speech models, and also lead innovation in speech and conversational AI solutions at Uniphore, focusing on end-to-end speech recognition, personalization, and related areas. My responsibilities span innovation, engineering, and taking projects to production. I am passionate about all aspects of Agentic AI, speech and language processing.