Jobs at Prime Intellect

Build post-training and agent systems for public-sector and regulated environments. Develop private improvement loops, permissioned tool-using agents, high-assurance evaluations, and deployable model updates within strict data and infrastructure boundaries. Conduct experiments, inspect traces, establish meaningful baselines, attribute failures, and collaborate directly with customers and research partners. The role requires hands-on model-training or agent experimentation, strong Python skills, sound research judgment, and comfort solving ambiguous problems.

Build environments, agents, evaluations, reward systems, and post-training methods for quantitative research and systematic decision-making. Develop realistic market and portfolio simulations, train long-horizon research agents, assess out-of-sample robustness, and create production feedback loops with monitoring and failure attribution. Collaborate directly with customers and research partners while maintaining a substantive research agenda.

Build environments, evaluations, agent systems, and post-training methods for embodied robotics in collaboration with an external robotics team. Train and evaluate agents using reinforcement learning, imitation learning, world modeling, planning, simulation, and multimodal perception. Own long-horizon agents, sim-to-real evaluations, reward systems, and training loops using offline and online feedback while accounting for safety and hardware constraints.

Build post-training and evaluation systems for multimodal computer-use agents operating browsers, desktops, and cross-application workflows. Develop long-horizon environments, reward signals, benchmarks, recovery strategies, and secure sandboxes using real workflows and production traces. Run experiments, inspect failures, attribute performance gains, and deploy successful research. The role involves external collaboration with customers, users, and research partners, requiring strong Python skills and ownership of ambiguous research problems.

Own externally facing applied reinforcement learning programs from ambiguous customer problems through deployment. Responsibilities include defining environments, evaluations, baselines, and reward signals; running post-training and distillation experiments; building agents with tool use and long-horizon capabilities; diagnosing model, data, harness, inference, and infrastructure failures; measuring customer impact; and converting successful work into reusable research, open-source systems, and platform capabilities.

Build agentic reinforcement-learning systems for semiconductor design and verification workflows. Develop interactive EDA environments, verifiers, reward functions, evaluations, and post-training experiments. Agents will generate or debug RTL, invoke simulation and formal tools, interpret failures, and optimize for correctness, efficiency, power, performance, and area. The role involves externally facing research, experimentation, failure analysis, and collaboration with customers, users, and research partners.

Conduct applied research on reinforcement learning for long-horizon, tool-using, multi-agent systems. Develop agentic judges, reward and credit-assignment methods, environments, evaluations, and scalable experimental systems. Diagnose reward hacking, data leakage, and infrastructure artifacts through rigorous experimentation. Build reusable open-source systems and publish findings while collaborating with customers, application teams, research partners, and open-source users.

Build continual learning systems that convert agent trajectories, corrections, and failures into training signals and model updates. Develop and evaluate streaming reinforcement learning, post-training, distillation, memory, and adaptation methods. Create reproducible task-generation, training, evaluation, regression, and rollback pipelines for changing task distributions. Conduct field experiments with external collaborators, assess retention and transfer, and establish promotion gates and human oversight for continuously improving production agents.

Own the development of open-source reinforcement learning, evaluation, agent, and environment systems. Build researcher-facing platform features, stable APIs, reference implementations, and technical documentation. Collaborate with engineers, customers, research partners, and external contributors to turn experimental research into reliable products and community contributions. Conduct model-training and post-training experiments, inspect traces, establish meaningful baselines, attribute failures, and improve workflows for researchers building and deploying advanced AI systems.