Handshake - AI PhD Student Researcher

• Shape how every career evolves in the AI economy, at global scale, with impact your friends, family and peers can see and feel • Work hand-in-hand with world-class AI labs, Fortune 500 partners and the world’s top educational institutions • Build a massive, fast-growing business with billions in revenue • Handshake AI builds the data engines that power the next generation of large language models. Our research team works at the intersection of cutting-edge model post-training, rigorous evaluation, and data efficiency. Join us for a Student Researcher engagement this fall where your work can ship directly into our production stack and become a publishable research contribution. The role can take place full time in person in San Francisco or potentially part time remote.) The target window of time is September - December 2026. • Projects You Could Tackle • LLM Post-Training: Novel RLHF / GRPO pipelines, instruction-following refinements, reasoning-trace supervision. • LLM Post-Training • LLM Evaluation: New multilingual, long-horizon, or domain-specific benchmarks; automatic vs. human preference studies; robustness diagnostics. • LLM Evaluation • Data Efficiency: Active-learning loops, data value estimation, synthetic data generation, and low-resource fine-tuning strategies. • Data Efficiency • Each intern owns a scoped research project, mentored by a senior scientist, with the explicit goal of an archive-ready manuscript or top-tier conference submission. • Desired Capabilities • Current PhD student in CS, ML, NLP, or related field. • Publication track record at top venues (NeurIPS, ICML, ACL, EMNLP, ICLR, etc.). • Hands-on experience training and experimenting with LLMs (e.g., PyTorch, JAX, DeepSpeed, distributed training stacks). • Strong empirical rigor and a passion for open-ended AI questions. • Extra Credit • Prior work on RLHF, evaluation tooling, or data selection methods. • Contributions to open-source LLM frameworks. • Public speaking or teaching experience (we often host internal reading groups).