AI Engineer

• You'll shape how the company builds. The harnesses and evals you design become the foundation every product runs on.
• You'll work on genuinely open problems. Making autonomous coding reliable, and building benchmarks where none exist, are problems the whole industry is still figuring out.
• You'll own outcomes end to end. From design to a production system, with real influence over technical and product direction.
• You'll have the tools. Claude Code is our primary environment, and Codex, Cursor Agent and similar tools are welcome. We provide compute for experiments.
• Credentials don't matter here. We care about what you can build and how you think, not where you studied or worked.

• Coding-agent harnesses. Design the environments agents work in: context packaging, tools, permissions, checkpoints, task delegation, resumability, traces, independent review and automated validation.
• Specialized eval harnesses. Build domain benchmarks from scratch: gold cases, hard negatives, adversarial scenarios, grading logic, human calibration, trajectory checks and regression gates that catch problems before customers do.
• Enterprise AI products. Turn customer needs into dependable Document AI and agent products that combine LLMs with deterministic systems, from scope and design through production monitoring and iteration.
• Model work, where it pays off. Fine-tuning experiments, model comparisons, data curation and inference economics, when they make the product better.
• Applied research. When a new paper or technique looks promising, you'll dig into how it actually works, test it against simple baselines and decide whether it earns a place in the product.

TypeScript and Node.js on AWS and Google Cloud, with Claude Code as our primary coding-agent environment, plus Codex, Cursor and similar tools. We're not dogmatic: when a problem calls for a different language, framework or cloud, we use it.

You don't need to tick every box. If you have the core experience below and are excited to learn the rest, please apply. We'll help you grow into the deeper agent and evaluation work.

• Serious coding-agent use. You use coding agents like Claude Code, Codex or Cursor heavily in your own work, well beyond autocomplete, and you have opinions about what makes them reliable or unreliable.
• Software engineering experience. You've built and shipped production software, ideally in TypeScript and Node.js, and you're comfortable reviewing code you didn't write, including code written by agents. Strong engineers from other languages are very welcome.
• Hands-on experience building with LLMs. You've built something real with LLM APIs: an agent, a tool, an automated workflow or an internal app.
• Curiosity about whether AI actually works. You care whether outputs are correct, not just impressive in a demo, and you want to get better at testing and evaluating them.
• Ownership. You enjoy taking a problem from idea to production without needing every step spelled out.

• Designing agent harnesses: context management, tool use, permissions, checkpoints, resumability or tracing.
• Building evaluation sets or benchmarks: gold cases, grading rubrics, LLM-as-judge, regression testing.
• Fine-tuning experience (SFT, LoRA, QLoRA or similar) or training small transformer models.
• Reproducing or critically evaluating machine learning research.
• Cloud infrastructure, security or production monitoring for AI systems.
• Startup or early-stage team experience.

You write and speak English fluently and enjoy working with international colleagues, including Japanese stakeholders who speak English as an additional language. Careful listening and checking for shared understanding matter more to us than accent.

No particular degree, university, PhD, minimum years of experience, big-name employer, publications, Japanese fluency or personal GPU setup. You don't need to have trained a frontier model. What matters is demonstrable practical fundamentals and a track record of learning fast.

We're a very small team, so priorities shift as the product evolves, and we move quickly toward commercial deadlines. In return, you get short paths from idea to production, real influence over how we build, and teammates who value careful evidence over hype. We especially appreciate people who make everyone faster by building reusable tools and infrastructure.

If you're a software engineer or AI engineer excited about LLM agents, AI coding tools and making AI systems reliable, we'd love to hear from you, even if you don't check every box. We're hiring now and hope to have someone start as soon as possible.