Vision RL Research Engineer: Multimodal Data & Tasks

Anthropic Limited; New York (NY); On-site; USD 350,000 - 850,000; Full time

Benefits offered by this job

Competitive compensation

Equity donation matching

Generous vacation

Flexible working hours

Job summary

Anthropic Limited is seeking a Research Engineer for the Vision team to drive end-to-end data strategies for visual capabilities and RL environments. You will own task specs, data quality, and rewards design while coordinating with external vendors and cross-functional teams.

The role combines applied research with hands-on data work, requiring collaboration with pretraining, RL, and product groups to translate environments into real-world vision-language capabilities.

Qualifications

• Bachelor's degree or equivalent in a relevant field.
• Experience with ML, CV, and RL research or applied work.
• Ability to design data-driven RL environments and evals.

Responsibilities

• Own data strategy for vision capabilities end-to-end, from evals to scalable RL environments.
• Manage vendor relationships, specify tasks, evaluate data quality, and iterate on rewards.
• Develop QA frameworks to detect reward hacking and ensure environment quality at scale.
• Run generalization experiments to measure data strategy impact on multimodal capabilities.
• Collaborate with pretraining, RL, and product teams to align research with practical applications.

Skills

Machine Learning

Computer Vision

Reinforcement Learning

Data Curation

RL Environments

Software Engineering

Education

Bachelor's degree

Tools

Python

PyTorch

RL frameworks

Job description

Anthropic Limited is seeking a Research Engineer for the Vision team to drive end-to-end data strategies for visual capabilities and RL environments. You will own task specs, data quality, and rewards design while coordinating with external vendors and cross-functional teams.

The role combines applied research with hands-on data work, requiring collaboration with pretraining, RL, and product groups to translate environments into real-world vision-language capabilities.