Multimodal AI Scientist — Research & Evaluation

Adobe; San Jose (CA); On-site; USD 165,000 - 239,000; Full time

Job summary

Adobe is seeking a research-focused engineer to identify advances with potential to create differentiated product capabilities. You will own or contribute to applied research from problem formulation through prototype implementation and validation.

You will curate multimodal training data, design architectures, and build robust evaluation pipelines to measure quality, robustness, safety, latency, and product usefulness.

Qualifications

• Advanced degree in CS/ML/AI or equivalent practical experience.
• Hands-on experience with transformer, diffusion, or multimodal architectures.
• Strong Python and PyTorch skills; experience with research-quality model code.
• End-to-end model development including data preparation, training, evaluation, and failure analysis.

Responsibilities

• Identify research advances with potential to create differentiated Adobe product capabilities.
• Own or contribute to applied research from problem formulation and experimentation through prototype implementation and validation.
• Curate and analyze multimodal training data, including sampling, filtering, deduplication, synthetic-data generation, and quality measurement.
• Design and optimize multimodal architectures, including vision encoders, modality projectors, and cross-modal attention.
• Build evaluation pipelines covering model quality, robustness, safety, latency, scalability, and product usefulness.
• Diagnose model failures through ablation studies, error analysis, and data- and architecture-level experimentation.

Skills

Transformer architectures

Multimodal architectures

Python

Model development

Failure analysis

Communication

Education

Master’s or Ph.D. in CS/ML/AI or related field

Tools

PyTorch

Job description

Adobe is seeking a research-focused engineer to identify advances with potential to create differentiated product capabilities. You will own or contribute to applied research from problem formulation through prototype implementation and validation.

You will curate multimodal training data, design architectures, and build robust evaluation pipelines to measure quality, robustness, safety, latency, and product usefulness.