Prithvi worked on improving the stability and reliability of the meta-pytorch/tritonbench repository by addressing a regression in dynamic runner initialization. Using Python and backend development skills, Prithvi identified that changes in triton_op.py disrupted the correct passing of tb_args to the BenchmarkOperator, which affected dynamic benchmark execution. By restoring the proper initialization path, Prithvi ensured that dynamic_runner functionality remained robust and prevented further breakages. This targeted bug fix reduced debugging time for benchmarks and enhanced CI reliability. The work demonstrated strong code navigation and software engineering practices, with clear, reviewable changes that contributed to the overall health of the repository.