Research Scientist, Human Data & Evaluation

Who we are and what we do

Contra Labs is a human-centered AI lab focused on creative and multimodal outputs, where human taste defines the next generation of AI capabilities.

We build industry-leading creative preference datasets that power benchmarking, evaluation, and post-training for the world's leading AI models and applications. Our work helps define what "good" looks like across design, video, imagery, and beyond.

Built on top of Contra, the professional network for independent creatives, Contra Labs connects frontier AI labs with a global network of top creative talent. Together, we turn human judgment into the training and evaluation layer that enables AI models and tools to power the next generation of human creativity.

Why this role exists

Contra Labs is expanding its research work with frontier AI labs across evaluation, post-training data, benchmarking, and human-centered research.

We need a researcher who can take an ambiguous client or research question and independently turn it into a rigorous project: identify the right research direction, develop the methodology, write the proposal, run the work, analyze the results, and produce insights that shape what we or the client should do next.

This person should complement our existing project execution capacity by bringing deeper research ownership and the ability to generate net-new ideas and approaches.

What you'll own

Research direction & proposals

Translate ambiguous client questions into clear research questions, hypotheses, and study designs.

Develop research proposals and methodologies from scratch rather than relying on a predefined approach.

Identify promising new research directions and push the team toward ideas we have not already considered.

Help define what Contra Labs should measure, study, and build next.

Evaluation & data methodology

Design human-data collection strategies, evaluation protocols, rubrics, annotation schemes, and experiments.

Develop strong prompt strategies for eliciting, evaluating, and comparing model behavior.

Determine appropriate sampling, evaluator selection, calibration, quality-control, and reliability methods.

Design methodologies that work across creative and multimodal domains such as design, image, video, and agentic workflows.

Analysis & insight generation

Analyze qualitative and quantitative evaluation data.

Move beyond reporting results to explain why patterns occur and what they imply.

Synthesize findings into clear recommendations, research narratives, and new hypotheses.

Produce client-ready reports, presentations, benchmarks, and research publications.