Platform / Infra Engineer

Emulated builds high-quality, long-horizon RL environments and post-training data for frontier labs — customers include all five US frontier labs (OpenAI, Anthropic, DeepMind, etc.) plus top international players, with active engagements across them. What makes it different: unlike most data vendors that treat post-training data as a human-services problem (throw bodies at capability gaps), Emulated treats it as a technical and research question — the forward-looking vision is to scale superlinearly with talent by building a model capable of producing post-training data autonomously (a very "bitter-lesson-pilled" company). It's cash-flow positive, raised a sub-$1M pre-seed, and ~3 months later an angel who'd tracked their progress made an impromptu seed offer they accepted (to fund the expensive model-training direction). The team (~9–12) is elite: a published neuroscience research lead who optimized Anthropic's circuit-tracer library (mech-interp) and built the world's fastest DBSCAN; founders Sid and Joseph with AWS software backgrounds (distributed databases; network infrastructure) — one a former musician, the other a chemistry olympiad; plus ex-Google/DeepMind/NVIDIA and a Stanford shock-physics PhD who works with national labs and the Stanford supercomputer. They just released a benchmark, Auto Research (evaluating models' ability to iterate in an ML-research loop), which drove strong frontier-lab interest.