Research Engineer, Visual Knowledge Work

About the role

About Anthropic

Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.

About the role

We're looking for a research engineer who believes that visual and spatial reasoning are core to fully unlocking the capabilities of LLMs. On the Vision team, you'll own the end-to-end process of creating training data and RL environments targeting visual knowledge work: identifying long-horizon and vision-heavy tasks, building evals, designing rewards, and scaling data. This is a unique role that combines applied research with hands-on data work. It's also highly collaborative — you'll partner with external vendors, pretraining, RL, and product teams to make sure the environments you build translate into real-world knowledge work capabilities.

What you'll do:

- Own the data strategy for vision capabilities end-to-end, from building evals and scaling RL environments

- Manage technical relationships with external data vendors, including writing task specifications, evaluating visual data and annotation quality, and iterating on reward design

- Develop and improve QA frameworks that catch reward hacking and ensure environment quality at scale

- Run generalization experiments to measure how data strategy changes improve multimodal capabilities on held-out evaluations

- Partner with pretraining, RL, and product teams, and do the science that shows we’re all rowing in the same direction

You may be a good fit if you:

- Have 7+ years of ML, computer vision, and software engineering experience through industry, academia, or other projects

- Have experience with reinforcement learning, reward design, or training data curation for large language or vision-language models

- Are familiar with the architecture, training, and operation of large vision language models

- Are comfortable managing technical vendor relationships and iterating quickly on feedback

- Are results-oriented, with a bias towards flexibility and impact

- Care about the societal impacts of your work

Strong candidates may also have experience with:

- Designing evals or benchmarks for LLMs or vision language models

- Large-scale pretraining, SL, and RL on language models

- Deep learning research on images, video, or other modalities

- Developing complex agentic systems using LLMs

- Large-scale ETL and data pipeline development

Representative projects:

- Writing a vendor-facing specification for a new family of visual RL training tasks, then iterating with the vendor on coverage, quality, and reward design