Samyak Jhaveri

ArXiv preprint describing GEPA, a genetic-Pareto prompt-optimization approach that improves OpenACC pragma generation for small LLMs, increasing compilation success and functional GPU speedups.

Paper and public release introducing ACCeLLiuM: SFT dataset and fine-tuned LLMs for generating expert OpenACC directives, demonstrating large gains over base models on held-out tests.

Preprint presenting an RL framework (GRPO) paired with a claim-level evaluator (DocLens) to optimize completeness and factual grounding in long-form clinical note generation.

Conference paper describing a quantum-annealing-based approach to code clone detection by mapping AST subgraph isomorphism to a quadratic optimization problem and solving it on a D-Wave quantum annealer.

Preprint introducing ParBench, a benchmark for reliable evaluation of LLM parallel code translation; submitted to NeurIPS 2027.