Principal Research Engineer
Required skills
A proven track record training large-scale video generation models from scratch, across multiple nodes, at the scale of millions of hours of data. Deep experience with post-training techniques at scale: RLHF, GRPO, DPO, and the judgment to know when and how to apply them. A proven track record data quality and you are not reluctant to question it. The ability to think strategically about multi-year research direction and execute hands-on at the frontier of what the team is building. Strong cross-functional influence: you shape how teams work together without needing positional authority to do it. A leadership style grounded in technical involvement. You lead by example, inspire through craft, and communicate with clarity. Genuine hunger to unlock new capabilities and an obsession with shipping. You're not satisfied by research that stays in a notebook. You want it in the hands of users. Particularly relevant experience Having owned a major model generation or capability jump end-to-end, from training runs through to product deployment. Working across both pre-training and post-training stages on the same model family, with direct accountability for the outcomes of both. Experience operating at scale: large distributed training runs, significant compute budgets, and multi-million hour data pipelines. Applying alignment and fine-tuning techniques in a video or multimodal context, not just text. Experience with human feedback pipelines applied to generative video or audio. Leading or significantly influencing the technical direction of a research team while remaining hands-on.
Responsibilities
Own the end-to-end technical direction for offline video generation, spanning pre-training and post-training, resolving the artificial boundary between those two stages in service of shipping better models faster. Partner with research leadership and team leads to define a unified long-term roadmap, broken into achievable objectives, and drive execution against it. Identify the most critical technical gaps across the video generation pipeline and jump in to unblock them, whether that means architectural decisions, training stability, post-training alignment, or cross-team coordination. Increase the velocity at which research ships to product: accelerate problem-solving, improve research-to-production handoffs, and increase visibility of research output in partnership with PMs. Coach and elevate more junior researchers and engineers toward senior technical thinking and execution. Help shape team structure and refine processes to enable high-velocity, cohesive execution across research.
Seniority
L7
Domain
Video generation, AI, machine learning, generative models
