Day to day, that looks like: constructing RL environments around spreadsheets, documents, and professional workflows. Writing verification logic and reward functions. Working with domain experts to scope what a correct answer actually looks like in an LBO model or a clinical note. Some days it's engineering, some days it's closer to research. The common thread is that you're producing the ground truth that frontier models get measured against.

• Build RL environments across finance, healthcare, and legal domains.

• Assist in designing tasks with golden answers, calibrated rubrics, and programmatic reward signals.

• Write verification logic and reward functions that can distinguish good model outputs from bad ones.

• Work directly with domain experts (investment analysts, physicians, attorneys) to translate complex professional workflows into structured tasks.

• Prototype new approaches to evaluation, verification, and synthetic data generation.

• Practical experience building with LLMs: prompting, evaluation, and agentic harnesses. You've built things that actually run, not just notebooks.

• High agency and technically sharp. You don't wait for permission, specs, or a roadmap. You see what needs doing, figure out how, and get it done.

• Comfortable working across very different contexts. The job moves between engineering, evaluation design, and deep collaboration with domain experts often in the same day.

• You ship and iterate. Small team, no room for work that sits in review. Bias toward getting something working, learning from it, and improving it.

• You own problems end to end, from scoping with a domain expert through to a working environment. If you prefer clearly partitioned tickets, this probably isn't the right fit.

• Already using LLMs as part of how you build, not just as the thing you're building for.