Research Scientist - Video Understanding

Role overview Join a research-focused team advancing the state of the art in multimodal machine learning with a specialization in video understanding. This position centers on designing and training models capable of temporal reasoning across long-form visual content, contributing to core platform capabilities. The role blends fundamental research with applied impact, including the opportunity to publish work openly.

Responsibilities - Design, train, and evaluate multimodal models that integrate visual, textual, and temporal signals for video understanding tasks - Develop approaches to temporal reasoning that enable models to track events, causality, and long-range dependencies across video sequences - Advance representation learning techniques for video data, including self-supervised and foundation-model methods - Build and refine core model components that power production-facing platform features - Collaborate with engineering and applied research colleagues to move prototypes toward deployable systems - Prepare and publish research findings, contributing to the broader research community through open publications

Requirements - Demonstrated research experience in one or more of: multimodal learning, video understanding, representation learning, or temporal modeling - Strong proficiency with deep learning frameworks and the practical engineering skills needed to train large-scale models - Solid background in machine learning fundamentals, including optimization, evaluation methodology, and experimental design - Track record of research output such as publications at major venues, open-source contributions, or equivalent industrial research work - Ability to work collaboratively across research and engineering teams

Nice to have - Experience with large-scale video datasets and the associated data pipelines for training and evaluation - Familiarity with transformer architectures adapted for spatiotemporal or multimodal inputs - Prior contributions to open-source model releases or reproducible research artifacts