Every major AI lab is chasing the same milestone: AI that can do real research. This team is building it now, at startup speed. Our client is an early stage AI lab backed by a top tier VC and founded by people from leading AI labs and universities. They’re developing AI systems capable of doing research work, along with the model research underneath that makes it possible. The team is small and senior. Every hire has a say in the research direction, and what you build goes into a live system, not just a paper. Location: Hybrid in San Francisco or New York preferred, with fully remote considered for the right person. Where you’d focus Improving how models reason Designing evaluations that show whether a model actually improved Teaching large language models through reinforcement learning Getting more performance out of models at scale Making models stronger and more dependable through better training Requirements 5+ years doing ML research yourself, not only managing it A Master’s or PhD in machine learning, computer science or a related field Publications at venues like NeurIPS, ICML or ICLR Research you’ve seen adopted in a product or production system Strong experience with PyTorch or JAX, including training at scale Preferred Hands on work with RLHF, post training or LLM evaluation Startup or early stage company experience Real world experience beyond university or large company research labs Why it’s worth a conversation An early seat on one of the most ambitious goals in AI A say in the research direction from day one Compute to match the ambition, plus meaningful equity Teammates who check results carefully and still move quickly Best suited to people who Want more scope than a large company or university lab can offer Have shipped in a fast moving, small team environment Like seeing their research used, not just published About this search: We’re running this search exclusively and on a retained basis, working as an embedded part of the client’s hiring team. All candidates for these roles come through us, and the client’s name is shared after an initial conversation.