Projects

Summer 2026 work at Lawrence Livermore National Laboratory with Tal Ben-Nun and Brian Van Essen. Integrated ring attention into OpenFold, splitting long-sequence attention across a ring of GPUs so no single device holds the full attention matrix, with a Triton kernel implementation. This made full-length structure prediction of Human Nebulin (NEBU_HUMAN, 8,525 residues) possible where baseline OpenFold runs out of memory.

Leading the development of a comprehensive validation and verification testsuite for the OpenACC Programming Model. This project ensures compiler compliance and reliability across diverse GPU architectures, supporting the broader HPC community.

Contributing to a predictive ecosystem for HPC software sustainability. Our CI/CD pipeline for LLVM Clang and new-Flang's OpenMP Offloading implementation runs comprehensive test suites and benchmarks on cutting-edge GPU hardware including NVIDIA H100, GH200, and AMD MI210/MI300A.

Developing novel approaches to leverage Large Language Models for automated test generation and validation in compiler verification workflows. This research explores the intersection of AI and systems software to improve testing efficiency and coverage.

Comprehensive continuous integration and deployment framework for validating OpenMP GPU offloading implementations. Focuses on automated testing, performance benchmarking, and cross-platform compatibility across diverse GPU architectures.

Developing unified testing methodologies for GPU computing across NVIDIA (H100, GH200) and AMD (MI300A) architectures. Ensures consistent performance and reliability for HPC applications across diverse hardware platforms.