Building inference systems for interactive AI.

Amey Agrawal

I recently completed my CS PhD at Georgia Tech, advised by Prof. Alexey Tumanov. My work co-designs scheduling, parallelism, memory, communication, simulation, and emulation across the inference stack. Previously at Microsoft Research, Azure Systems Research, and Qubole.

Current work. Low-latency streaming inference across systems, workloads, and hardware.

EmailCVGoogle ScholarGitHub

$ selected research

Selected work across scheduling, capacity planning, long-context serving, and emulation.

Long-context serving SoCC'26

Medha

Long-context serving across scheduling, parallelism, and memory

Paper
Emulation

Revati

GPU-free time-warp emulation for reproducible serving experiments

Paper

$ all publications

::

Akasha: Breaking the I/O Wall in Hierarchical KV Cache Management

Amey Agrawal*, Elton Pinto*, Souradeep Bera*, Sukrit Kumar, Irene Paul Stephen, Chus Antonanzas, Anirudha Agrawal, Aldinash Seitenov, Prakhar Jagwani, Anmol Agarwal, Haoran Qiu, Alexey Tumanov
::

Setu: A Global Tensor Exchange for Composable Transfers in Large-Scale ML Systems

Elton Pinto*, Amey Agrawal*, Aldinash Seitenov, Anirudha Agrawal, Bhumika Chopra, Roshan Dathathri, Sadjad Fouladi, Alexey Tumanov