Research Scientist, Video Understanding

Location: New York, on-site

Salary: $200K–$250K annually, plus equity

Company

hirify.global builds data infrastructure and high-quality real-world datasets for robotics and embodied AI.

What you will do

• Own the video understanding research agenda across manipulation, locomotion, daily activity, and long-horizon behavior.
• Design model architectures and training strategies for large-scale video representation and video-language models.
• Run self-supervised and multimodal pretraining with rigorous evaluations, clean ablations, and regression tracking.
• Train and fine-tune video encoders, temporal transformers, joint-embedding models, and multimodal models.
• Incorporate pose, depth, camera motion, and optical flow priors when they improve representation quality.
• Convert model checkpoints into production-ready embeddings and outputs for retrieval, labeling, QA, and analytics.

Requirements

• Deep experience training large models in PyTorch or an equivalent framework, including multi-GPU or distributed training.
• Strong understanding of modern video representation learning and/or multimodal modeling.
• Experience with large-scale video encoders, video-language models, VLMs, VLAs, or temporal representation learning.
• Ability to run rigorous experiments and communicate results clearly.
• Strong software engineering discipline and ability to write research code that can be shipped.
• Apply to only one Research Scientist position; additional applications may result in rejection of all research submissions.

Nice to have

• Experience with video VLM or VLA-adjacent systems such as VideoCLIP, InstructBLIP-Video, or LLaVA-Video-class models.
• Experience with egocentric or embodied datasets such as Ego4D, EgoExo4D, EPIC-Kitchens, or Something-Something.

Culture & Benefits

• High standards for technical excellence and first-principles reasoning.
• Truth-seeking culture with rigorous measurement and experimentation.
• High-intensity environment with rapid execution and end-to-end ownership.
• Work on scarce egocentric embodied video data and research that directly feeds production systems.
• Equity offered in addition to salary.