Ashwin Sreedhar

LLM inference systems and the data around them. I build emberserve, a from-scratch inference engine with paged KV cache, prefix caching, continuous batching and an OpenAI-compatible server, benchmarked against vLLM on A100s and Runpod Serverless. serverless-lakehouse is the PySpark + Delta medallion pipeline over those benchmarks — cold starts, FlashBoot, worker boot anatomy, cost per request — with the gold layer served as a Space here. MS Computer Engineering, Purdue (Dec 2026); previously backend at Gallo.