Sara Mahdizadeh Shahri


I am a Ph.D. student in the Electrical and Computer Engineering Department at Carnegie Mellon University, advised by Prof. Akshitha Sriraman. My work bridges software systems and computer architecture to enable efficient hyperscale web services while improving user experience across all user demographic groups.

Data center systems have traditionally been designed with a “performance-first” approach. My dissertation shows that this approach can inadvertently degrade the experience of certain user groups. Web schedulers and vector databases adopt optimizations such as request prioritization and response approximation to reduce average latency, and these optimizations produce far worse responses for users in the tail of the distribution. My work identifies where these disparities arise, introduces metrics to measure them, and builds cross-stack systems to mitigate them. I am currently extending this line of work to retrieval-augmented chatbots, where the prefill stage dictates both the cost of serving a request and the latency the user actually feels.

I am also a Student Researcher with the Systems Research Group (SRG) at Google. Before CMU, I spent a year as a Ph.D. student at the University of Michigan, where I worked with Prof. Baris Kasikci on front-end bottlenecks in data center applications. I received my M.Sc. from Penn State under Dr. Aasheesh Kolli, where I worked on computer architecture and persistent memory, and my B.Sc. from Sharif University of Technology. I interned at Google in the summers of 2021 and 2022.

My research is supported by the Benjamin Garver Lamme/Westinghouse Graduate Fellowship, and has previously been supported by the CyLab Presidential Fellowship, the K&L Gates Presidential Fellowship, and the CMU College of Engineering Presidential Fellowship.
See my work

Publications