Research Scientist, Video & Multimodal
Core
Design, structure, and evaluate video and multimodal data to advance video understanding, temporal reasoning, and video generation models for frontier labs and AI builders.
Role type
Senior IC research scientist (video & multimodal)
Builds
Evaluation methodologies, data specifications, and benchmarks for video and multimodal AI systems
Domain
Artificial Intelligence, Computer Vision, Video Processing
Deliverable
production ML models
Required skills
video understanding, temporal reasoning, multimodal model training, PyTorch, dataset design, evaluation methodology, adversarial testing, open-source contributions
Preferred skills
responsible AI evaluation, red-teaming, synthetic data generation
Technologies
PyTorch, ffmpeg, decord, WebDataset, Parquet, Arrow, HuggingFace, transformers, PEFT
Responsibilities
Define data specifications for video understanding and generation tasks; build evaluation methodologies for temporal grounding and fidelity; run experiments to validate data decisions; design adversarial evaluations; publish benchmarks and papers; collaborate with annotation teams and customers.
Seniority
Senior, hands-on IC
