Research Scientist Senior​/Principal

Research Scientist / Principal Research Scientist - Multimodal VLMs & Video Understanding

Interested in working on Multimodal VLMs, long-form video understanding and post-training where the research has a direct path into production?

This is a hands‑on research role within a small team building models for complex real‑world video understanding.

You'll work across training, post‑training, evaluation and deployment, with a particular focus on long‑context video, temporal reasoning and improving model performance through SFT, reinforcement learning and other post‑training approaches.

• Training and improving vision‑language and multimodal models.
• Developing post‑training pipelines using SFT, RL and parameter‑efficient tuning approaches.
• Working on long‑context video understanding, temporal reasoning and memory.
• Designing new training objectives and experimentation strategies for fine‑grained perception and reasoning.
• Building evaluation frameworks for video understanding, activity recognition and real‑world model performance.
• Exploring distillation, quantisation and model routing for efficient deployment.

At Principal level, you'll also help shape the research roadmap, set technical direction and mentor other researchers.

You'll have a strong background in deep learning, ideally across Computer Vision, multimodal learning or video understanding, alongside hands‑on experience training transformer‑based models.

Experience with VLM post‑training, RL, long‑form video or large‑scale model development would be particularly relevant. A strong publication record is useful, but practical research impact is equally important.

The team has access to a large proprietary video dataset and dedicated compute, allowing researchers to test new training and post‑training approaches on problems that are difficult to reproduce in a traditional lab setting.

If you're looking for a research role where you can push VLM and video models further through post‑training, while still seeing the results used in production, we'd love to hear from you.