Multimodal AI Research Scientist — Production‑Driven

NEPSE Trading; San Francisco (CA); Remote; USD 197,000 - 263,000; Full time

Job summary

Descript is hiring an Applied Research Scientist focused on multimodal understanding for AI editing features. You’ll design and evaluate vision-language systems, build datasets (synthetic when needed), and train models from scratch or fine-tune foundation models.

You’ll ship production-ready features with engineers, and may pursue publications. The role emphasizes rapid experimentation, rigorous evaluation, and clear communication about progress and roadblocks.

Qualifications

• Proven ability to design and implement deep learning algorithms.
• Strong programming skills with PyTorch and Python.
• Track record of publishing or shipping model-based work.

Responsibilities

• Multimodal understanding: build vision-language systems for editoric features.
• Design benchmarks and evals for editorial quality vs. cost and latency.
• Create datasets and synthetic data when real data is scarce.
• Train models from scratch or fine-tune foundation models.
• Ship models from prototype to production with cross-functional teams.
• Identify next research directions to become product features.
• Publish work at academic venues if desired.

Skills

Deep learning

PyTorch

Experimentation

Communication

Research direction

Python

Education

PhD or Master in deep learning or related field

Equivalent experience

Tools

Git

Job description

Descript is hiring an Applied Research Scientist focused on multimodal understanding for AI editing features. You’ll design and evaluate vision-language systems, build datasets (synthetic when needed), and train models from scratch or fine-tune foundation models.

You’ll ship production-ready features with engineers, and may pursue publications. The role emphasizes rapid experimentation, rigorous evaluation, and clear communication about progress and roadblocks.