Bio

I’m a Sr. SDE @ AMD, where I work with the CodeGen team (Previously, Nod.AI) remotely. Previously, I was a grad student @ Department of Computer Science and Automation , IISc Bangalore, where I was advised by Prof Arkaprava Basu and worked at CSL. My research was primarily focussed in Computer Systems(Architecture, Operating Systems & GPU Memory Subsystem).

At AMD, I’m focusing on high-performance compiler technology for advanced GPU architectures. My work spans system software optimization, dynamic dispatch for modern DNN workloads, and cross-stack integration with PyTorch and MLIR. Most of my work is on open source repositories like iree, torch-mlir, llvm.

Previosly at Cerebras, my work revolved around increasing usability and performance of systems for AI training/Inference. To this end, I’ve worked with MLIR/LLVM, torch-mlir, and python, amongst a lot of other frameworks. I also worked at compiler support for the underlying architecture at Cerebras, which involves multiple layers of tiered dialects and abstractions, from python, all the way down to the low-level IR supported by the chip.

At IISc, My thesis was about accelerating pre-processing pipelines for Deep Neural Network Training tasks to alleviate system bottlenecks (CPU / Caches / Memory / Disk) and attaining a speedup in end-to-end training. I built a Library on top of DALI that automatically determines the underlying system properties for potential speedup opportunities, and builds pre-processing pipelines that are upto 10.5x faster that vanilla DALI pipelines. With these accelerated pre-processing pipelines, I was able to attain upto 67% end-to-end training with an H100 GPU.

Prior to joining IISc, I interned at major tech startups like Innovaccer and Atlan between 2019-2020 where I primrarily worked on ETL of data; from warehouses to UI based data-products. I used technologies like React, Django, Flutter, PostgreSQL, Apache Airflow to achieve such tasks.

My undergrad degree is from USICT, GGSIPU (Dwarka) where I majored in Electronics and Communications Engineering from 2017-2021

EMAIL : k e s h a v j h a [at] i i s c [dot] a c [dot] i n