Back to jobs
Remy logo

Head of AI Research

RemySan Francisco

Posted November 17, 2026

Job Description

About Remy

Remy is an AI agent tackling one of the most exciting frontiers in creative AI: autonomous product design. More than just generating UI, Remy helps software teams visually explore, refine, and complete verified design outcomes.

The old way is slow: text-heavy PRDs, bottlenecked design cycles, weeks of back-and-forth before a team agrees on what to ship. Remy uses generative AI to align teams visually from day one, learning their style, standards, and goals to produce design outputs ready for use by designers, engineers, and coding agents.

As Remy integrates deeper into enterprise workflows, each interaction makes it smarter. Our vision is a fully autonomous AI product team that can ideate, design, and ship software on its own.

You'll be joining as our Head of AI Research.

You'll own the research behind Remy's design taste: how we measure whether a design is good, how we train models and agents to make better ones, and how every user interaction improves the next output. You'll work side-by-side with our CEO and Engineering, build the research team, and decide what we work on and what we stop.

This is a ground-floor role with massive ownership. The research you run becomes the product.

About the role

  • Set the research agenda. Pick the bets that move design quality, run them, and kill the ones that don't pan out. Report results to the team in numbers.

  • Make taste measurable. Build reward models, LLM and VLM judges, and preference datasets that agree with professional designers (or model their disagreements). Keep calibrating them against human labels as the product changes.

  • Post-train for design quality. Use RL, preference optimization, and test-time search so our agents make designs that are better and more varied, without collapsing into generic AI output.

  • Learn from every user. Turn feedback, edits, and picks into training signal so Remy learns each team's style, standards, and goals.

  • Ship it. Take research into production with our engineers and measure it against real usage. Publish and open-source when it helps.

  • Build the team. Hire and mentor researchers and research engineers, and set the bar for how we run experiments.

Tech stack

Python • Modal • TypeScript • HuggingFace • Vercel AI SDK • AWS • Docker

About you

  • A body of research you can point to. First-author papers at venues like NeurIPS, ICLR, CVPR, CHI, or SIGGRAPH, or open-source work and technical writing that others build on. We care about the work more than the venue.

  • Depth in at least one area we need. Reward modeling and preference learning, RL post-training for LLMs or diffusion models, evals and LLM-as-judge methods, generative models for images, layout, or UI, or human-centered AI for creative tools.

  • A genuine eye for design. You have training or a practice in design, art, architecture, photography, film, or music, and you can explain why one output beats another. Your taste shows up in your research.

  • Research that shipped. You've taken a model or method from experiment to a product real users touched.

  • Experience leading people. As a research lead, tech lead, or founder, you've hired, mentored, and set direction for others.

  • Rigor about measurement. You know how judges and benchmarks fail, including position bias, reward hacking, and mode collapse, and you design experiments that catch it.

  • Clear, precise communicator. You can write a crisp research proposal, explain a negative result, and disagree constructively.

Bonus points for:

  • Building human-preference datasets with expert annotators, especially designers.

  • Personalization, such as modeling an individual's or a team's taste.

  • Long-running or self-improving agents, “autoresearch” or RSI, prompt and program optimization, or test-time training.

  • Research on diversity and novelty in generation, or on detecting and training out AI slop.

  • Early-stage startup or founder experience.

  • A PhD in ML, HCI, computer vision, or a related field. Not required if your work speaks for itself.

If you want to define what "good design" means to a model, prove it with numbers, and ship it to the teams deciding what software gets built, we'd love to meet you!

<aside>
✅

Please reach out to [email protected] to apply or with any questions.

</aside>