Rajveer Bachkaniwala is a researcher and computer science graduate student focused on making systems for serving large language models faster and more efficient. His work explores practical ways to improve inference performance and identify bottlenecks in machine learning workflows. His projects include Stream2LLM, which overlaps context streaming with LLM prefill and reports up to 11 times faster time to first token.

The work is associated with MLSys ’26 and has a project page. Another project, Lotus, profiles machine learning data-preprocessing pipelines to uncover GPU utilization bottlenecks. Lotus is associated with IISWC ’24 and HotInfra ’24 and is also presented on a project page.

Bachkaniwala is pursuing an MS in Computer Science at Georgia Tech, with graduation planned for May 2026. Before that, he spent four years in Georgia Tech’s PhD program, and he earned a BS from Penn State. His website brings together information about his research, publications, and blog, alongside links to GitHub, X, and LinkedIn