No open roles yet. Come anyway.

The work spans reinforcement learning, computational topology, world models, robotics simulation, and eventually hardware. Nobody has all of it. We care about depth in at least one of these and curiosity about the rest.

• 01Exploration and RL research. Go-Explore, RND, ICM, archive-based methods, curriculum generation, and the patience to run honest baselines at benchmark scale.
• 02Computational topology and geometry. Persistent homology, cubical and witness complexes, Ollivier–Ricci curvature, spectral methods, and making them fast enough to run inside a training loop.
• 03Robotics simulation. Isaac Sim, MuJoCo, Omniverse, procedural 3D environment generation, sim-to-sim transfer, and the plumbing to run all of it on a cluster.
• 04World models and representation learning. Latent spaces an agent can build a map over, and the signals that reveal gaps in a model's own knowledge.
• 05Systems and infrastructure. Ray, PyTorch, GPU pipelines, reproducible-to-the-seed benchmarks, and open-source library craft.

Everything we ship is deterministic up to seeds. Every world in TopoGym carries its exact shape as the answer key, so a claim about exploration is something you can check, not something you have to believe. We publish, we open-source the environments, and we run the boring baselines all the way out.

We are remote-first with a base in New York, and we work closely with academic collaborators. Early hires will shape the research agenda, the platform, and the company itself.