AI Engineer

🎯 The role Our research team works on the core problems behind that mission: how agents learn in these environments, how we measure what they can do, and how we turn that into better training data and training methods. As a Research Engineer, you build and scale the systems that research runs on. That includes post-training and reinforcement learning pipelines, evaluation harnesses, sandboxed environments and data systems. You also run the experiments yourself. You take ideas from a first prototype to something the rest of the company and our partners rely on. This is a hands-on individual-contributor role. You're not measured by tickets closed. You're measured by technical leverage: stronger systems, faster experiments and research outcomes that change what we can do. 🚀 What you'll work on • Build and run training and post-training pipelines, including supervised fine-tuning, reinforcement learning and preference optimisation, and scale them across GPU clusters. • Build evaluation systems, graders and feedback loops that tell us what agents can do, where they fail and where the headroom is. • Build and scale the sandboxed, resettable environments that agents act in, and the rollout infrastructure that connects them to training. • Build systems that generate, filter and curate high-quality tasks, trajectories and synthetic data at scale. • Own research problems from hypothesis through experiment, evaluation and iteration to production. • Make experiments fast, reproducible and trustworthy, so the whole team can move quicker and believe its results. • Work across Research, Platform and delivery teams on problems that don't fit neatly into research or engineering. • Set technical direction on the work you lead, mentor others, and raise the engineering bar for research code. 🧠 What we're looking for • Hands-on experience training, fine-tuning or post-training models, or building the evaluation and agent systems that sit around them. You should have done this yourself, not only consumed model APIs. • Strong software engineering, particularly in Python. You write clean, testable code and build systems that other people depend on. • Solid ML fundamentals, and the judgement to know whether an intervention actually worked, not just whether a number went up. • Experience designing experiments, choosing sensible success criteria and making decisions from noisy or ambiguous results. • Ownership of substantial technical work over months, not only short experiments or isolated features. • Comfort turning open-ended problems into concrete plans without a playbook to follow. • A pragmatic, low-ego approach. You challenge assumptions, change direction when the evidence says so, and optimise for impact over novelty. ✨ Useful, but not essential • Experience with LLMs or multimodal foundation models, including distributed training (for example PyTorch, Ray, DeepSpeed, vLLM). • RL or post-training in practice: RLHF, DPO, GRPO, reward models or rollout systems. • Agent harnesses, tool use, sandboxed code execution or long-horizon agent evaluation. • Large-scale training data, synthetic data or data-curation systems. • Taking research from prototype to production in an early-stage or fast-scaling company. • Leading multi-person technical work while remaining an IC. • Public technical work: open-source projects, models, datasets or papers. Academic credentials are not required. This role is not a prompt-engineering or LLM API integration role, a people-management role (former managers who want to build again are welcome), or a pure research role without ownership of working systems.