InfrastructureEngineer

About Bespoke Labs Bespoke Labs is an applied AI research lab pioneering data and RL environment curation for training and evaluating agents. Recently, we curated Open Thoughts, one of the best open reasoning datasets used by multiple frontier labs, trained SOTA specialized models such as Bespoke-MiniChart-7B and Bespoke-MiniCheck, and built the environment infrastructure that frontier labs and enterprises use to make their agents reliable. Bespoke is uniquely positioned to capture a large share of data and RL environment curation. About the Role We're looking for an Infrastructure Engineer to own the execution layer beneath our RL environments: the systems that let an agent operate inside a realistic, multi-tool world coherently for hours or days. This is a hard systems problem disguised as an AI job. As the tasks agents can complete keep lengthening, the environments that train them have to stay coherent across far longer horizons than anything that exists today. That means sandboxing and isolation you can trust, execution that's fast and cheap enough to run at training scale, and the ability to snapshot, restore, inspect, and branch a running environment instead of treating every rollout as one-shot. You'll build the platform that makes all of this possible. You'll work closely with our research and data teams, and directly with frontier labs and enterprise customers, to turn environment designs into infrastructure that runs reliably in production. What You'll Do 1. Environment Execution & Sandboxing: - Design and own the sandboxing and execution layer that environments run inside. Build systems to snapshot and restore environment state (disk, process, and where relevant memory and accelerator state) so runs can be paused, resumed, inspected, and branched rather than executed once. - Develop the machinery to detect failure modes early in a rollout (reward hacks, infra faults, fairness issues) and to revert to a known-good state, patch, and continue. - Extend exec