Senior Research Scientist, Video & Multimodal AI

Innodata Inc.; United States; Remote; USD 160,000 - 185,000; Full time

Job summary

Innodata Inc. is hiring a Research Scientist to own the science behind video evaluation for multimodal models. You will partner with customers and frontier labs to design data specifications, annotation schemas, and evaluation criteria.

You will run experiments proving data decisions matter, publish benchmarks, and turn findings into reproducible workflows with clear documentation for partners. Collaboration with annotation teams and researchers is essential to drive meaningful datasets and

Qualifications

• 5+ years hands-on industry experience in video understanding or multimodal ML.
• Bachelor's degree in CS/EE required; advanced degree preferred.
• Strong PyTorch fundamentals and experience with video datasets.

Responsibilities

• Translate requirements of video and multimodal models into concrete data specifications and evaluation criteria.
• Build evaluation methodology for video understanding and generation, including temporal grounding and cross-modal tasks.
• Run experiments to prove data decisions matter and publish findings.
• Collaborate with annotation teams and customers to shape data collection and labeling plans.
• Document reproducible experiments and present results to researchers and partners.
• Develop benchmarks and papers that advance the field.

Skills

Video understanding

Multimodal ML

Temporal reasoning

Long-form video

Cross-modal grounding

Experiment design

Benchmarks

Education

Bachelor's degree in CS/EE

Advanced degree preferred

Tools

PyTorch

FFmpeg

Decord

WebDataset

Parquet

Arrow

HuggingFace datasets

HuggingFace transformers

PEFT

Job description

Innodata Inc. is hiring a Research Scientist to own the science behind video evaluation for multimodal models. You will partner with customers and frontier labs to design data specifications, annotation schemas, and evaluation criteria.

You will run experiments proving data decisions matter, publish benchmarks, and turn findings into reproducible workflows with clear documentation for partners. Collaboration with annotation teams and researchers is essential to drive meaningful datasets and