About Me
I’m a third-year PhD student in Computer Science at Georgia Tech, advised by Ada Gavrilovska in the Kernel Lab. I work at the two-way intersection of systems and machine learning — building systems for AI and bringing AI into systems — with a throughline of storage and memory hierarchies.
My current work builds AI runtime systems that let multi-agent LLM workflows for scientific computing run efficiently on high-performance computing platforms. Earlier, at Carnegie Mellon’s Parallel Data Lab, I worked with Greg Ganger and Rashmi Vinayak on efficiently managing data across redundancy schemes in cluster storage. At Microsoft Research, I collaborated with Sid Sen, Chetan Bansal, and Gagan Somashekar on AI-driven methods for cloud reliability testing.
I can be reached by emailing dax [at] gatech [dot] edu.
Highlights
A prediction-driven prefetch runtime that sits between an LLM agent and the HPC resources it needs. After each LLM response it predicts which model weights or data files the next tool call will require, loads them speculatively in the background, and cancels if the agent diverges — overlapping expensive I/O with active computation instead of stalling on it.