Samyak Jhaveri

Final-Year Software Engineering PhD Candidate at UC Irvine and ex-Applied Research Intern at Oracle Health AI. My research specialization is in AI for Agentic Software Engineering for Parallel and Distributed Accelerated Computing Systems. Seeking Applied Scientist / Research Scientist roles where LLMs meet HPC, compilers, and accelerated computing, with interest in Finance, Health, and Energy. As a researcher, I developed domain-specific training pipelines to post-train models using Supervised Fine-Tuning, and RLVR/GRPO to write parallel, high-performance code. I also developed benchmarks and evaluation harness infrastructure to evaluate LLM and coding agents' capabilities in HPC parallel and distributed code translation. I have experience across the full AI research workflow: hypothesis, experimental design, data, training, evaluation infrastructure, and write-up. Available 2027. STEM OPT, three years of work authorization. Specialties: LLM Post-Training, Supervised Fine-Tuning, GRPO, RLVR, Reward Design, LLM Evaluation & Benchmarking, Prompt Optimization, CUDA, OpenMP, OpenACC, GPU Programming, HPC. Contact: samyakjhaveri2799@gmail.com. One peer-reviewed paper to date (Onward! 2023, ACM SIGPLAN/SPLASH) and one published MS Thesis; other work is preprint or under review. As an Applied Research Scientist Intern at Oracle Health, I post-trained, tested, and deployed LLMs for enhancing clinical SOAP note generation from ambient doctor–patient dialogue using business-aligned reward modeling in an evaluation-integrated reinforcement learning pipeline. I built a GRPO/RLVR loop from scratch in one internship—including dataset creation, reward design, training, serving, and evaluation—using a claim-level evaluator directly as the reward, yielding a claim-level F1 improvement of +4.6 points over the base (78.8% vs. 75.3%); first-author preprint arXiv:2510.02338; patent pending. For my M.S. Thesis in Software Engineering, I developed a quantum computing algorithm for code clone detection by expressing it as a graph/subgraph isomorphism problem mapped to the QUBO framework for adiabatic quantum computing; this work generalizes to other graph comparison domains. MS Thesis: "Quantum Computing for Software Engineering: Code Clone Detection and Beyond" (https://escholarship.org/uc/item/4kq183hk#main). arxiv+2

Research at the intersection of software engineering and quantum computing: developed and published methods that express software problems (e.g., code clone detection) as optimization tasks solvable on quantum annealers. acm+2

Applied LLMs and tooling for high-performance and parallel computing: led projects (ACCeLLiuM) and prompt-optimization work that use LLMs to generate OpenACC pragmas and enable GPU offloading at scale. arxiv+1

Broader ML/AI applications in health and long-form generation: contributed to frameworks that optimize clinical-text generation for factual grounding and completeness using RL-based reward schemes. arxiv+1

Blends academic research with short-term industry and visiting research appointments: has held visiting researcher / intern roles (USC ISI, UC Irvine research positions) and presents work at conferences and seminars. splashcon+2