Research Engineer, Visual Knowledge Work

About this role

Anthropic seeks a Research Engineer to advance visual and spatial reasoning capabilities in Claude, their large language model. You'll conduct experiments on vision architectures and datasets, develop evaluation benchmarks, and collaborate with product teams to solve real-world customer challenges involving multimodal AI.

What you'll do

• Run experiments to test architectural variants, data strategies, and supervised/reinforcement learning techniques for vision improvement
• Develop tools and agentic infrastructure enabling Claude to reason over visual inputs
• Create evaluations and benchmarks measuring multimodal capabilities during training and deployment
• Partner with product organization to address vision-related API customer challenges
• Work on full-stack vision improvements across pretraining, RL, and runtime techniques

What they're looking for

• Computer vision and machine learning (7+ years industry/academia experience)
• Large vision language model architecture and training
• Visual dataset creation and evaluation (synthetic and real-world)
• Systematic prompting, finetuning, and evaluation
• Deep learning frameworks (PyTorch, JAX)
• Large-scale ML systems and data pipeline development
• Agentic system development with LLMs
• Collaborative problem-solving and pair programming

Benefits

• Competitive compensation range: $350,000–$850,000 USD annually
• Visa sponsorship available
• Hybrid work policy requiring 25% office presence (role-dependent)
• Collaborative research environment focused on safe, beneficial AI
• Full-stack projects with real-world product impact
• Work on cutting-edge multimodal AI research