Multimodal AI Research Scientist — Remote

Mosaec; San Francisco (CA); Hybrid; USD 197,000 - 263,000; Full time

Benefits offered by this job

Healthcare package

401k matching

Catered lunches

Flexible vacation time

Job summary

Descript is seeking a premier deep learning researcher to advance multimodal understanding for editor-focused AI tools. You will design vision-language systems that empower Descript’s agentic editing features, build benchmarks to quantify editorial quality, and create datasets including synthetic data for scale.

You’ll train models from scratch or fine-tune foundation models, collaborating with cross-functional teams to ship features that scale to millions of creators.

Qualifications

• Proven ability to design and implement deep learning algorithms, publications, open-source work, or shipped models.
• Strong programming skills with deep fluency in PyTorch.
• Track record of generating new ideas in machine learning and running multiple experiments quickly.

Responsibilities

• Multimodal understanding: build vision-language systems for Descript’s editing features.
• Design benchmarks and evals to measure editorial quality vs cost/latency.
• Build and curate datasets, including synthetic data when real data is scarce.
• Train models from scratch or fine-tune foundation models for required capabilities.
• Ship models from prototype to production with engineering teams.
• Identify future research directions that could become Descript features.

Skills

Deep learning

Programming in Python

Experimental design

Communication

Education

PhD or Master’s in deep learning

Tools

PyTorch

Job description

Descript is seeking a premier deep learning researcher to advance multimodal understanding for editor-focused AI tools. You will design vision-language systems that empower Descript’s agentic editing features, build benchmarks to quantify editorial quality, and create datasets including synthetic data for scale.

You’ll train models from scratch or fine-tune foundation models, collaborating with cross-functional teams to ship features that scale to millions of creators.