ML Research Team Lead

• Research Team Lead
• AI learned language. Now it needs to learn worlds. Origin Lab builds the data infrastructure for that shift: premium, rights-cleared multimodal content for the AI labs training world models. Licensed at the source. Structured for learning. Delivered with full provenance.

Backed by Lightspeed, SV Angel, and a deep bench of Silicon Valley investors and operators, including the founders of Twitch and Cruise. Small team. Moving fast. Every person shapes the business directly.

• The role
• The work is world models. You will manage our team of ML research engineers, three today, and recruit the rest. We hand you 10 modalities of ground truth, sourced and measured directly from game engines, and the mandate to prove what that data does that scraped video cannot.

This is a player-coach seat. You set the research agenda, run the operating cadence, and stay in the code. You read your team's training runs, reproduce their numbers, and diagnose the experiment that is quietly wrong.

Based in San Francisco, Los Angeles, or New York.

• What you'll own
• The research agenda. Which tracks get staffed, which get killed, and in what order. Publish the reasoning so the rest of the company can plan against it.
• The team. Hire, level, and grow research engineers. Define what a research week looks like and what done means for an experiment.
• The standard of evidence. Matched compute, preregistered endpoints and kill criteria, paired statistics, reported nulls. Kill criteria are set before the run, not after the plot.
• Publications and academic partnerships. A calendar of papers, open benchmarks, and released evaluation harnesses, and the research relationships that carry them. You represent the work externally.
• The translation layer. Turn customer and business needs into research questions, and tell sales and product what our data can and cannot support today.

• What we're looking for
• 10+ years hands-on in ML, building and training video models, ideally world models. We are looking first at people who did that work on a multimodal, video, or world-model team at a frontier lab: Google DeepMind, NVIDIA, Meta, OpenAI, Runway, Luma, or similar.
• Graduate degree in a relevant field, PhD strongly preferred, with first-author work at CVPR, NeurIPS, ICML, ICLR, or equivalent venues.
• Breadth across the fields world models touch: video, 3D and spatial, physical AI, robotics, autonomous vehicles.
• 3+ years leading researchers, and the network to recruit a world-class team.
• Rigorous experimental design and clear technical writing.

• Nice to have
• You have worked with video game data, whether that is capture, engine telemetry, modding, or game-derived training sets. This is the largest plus on the list. Our corpus is engine-native, and someone who already knows what comes out of a game engine starts a quarter ahead.
• You have trained or evaluated models on the problems our data is built for: metric depth and camera pose, camera-conditioned and controllable video generation, action-conditioned prediction. You have opinions about how these are measured and where the published benchmarks fall short.
• You have built a research function inside a small company rather than joining one that already ran. The experiment cadence, the evaluation infrastructure, and the publication process did not exist until you made them.