Applied Research - Agentic RL

Prime Intellect•San Francisco, CA

1d•$150,000 - $300,000

About The Position

Prime Intellect is hiring for a broad, cross-domain Applied Research role focused on reinforcement learning and ML systems for agents that reason over long horizons, use tools, coordinate with other agents, recover from failure, and improve from experience. The work spans algorithms, environments, harnesses, reward and credit-assignment methods, evaluation, and scalable experimental systems, with methods intended to transfer across application domains. The researcher will turn workflows into tasks, environments, evaluations, and reward signals; run post-training experiments against real baselines; and determine whether observed gains come from the model or from data, harnesses, inference, or infrastructure. The role extends Prime Intellect's existing work in areas including verifiers, prime-rl, multi-agent training, Prime Agent, and RL systems tested at trillion-parameter scale, and pressure-tests approaches against live application programs. Applied Research roles are externally facing. Researchers work directly with customers, users, or research partners while maintaining a substantive research agenda. Reusable results may become shared research, open-source systems, papers, and product capabilities.

Requirements

Relevant hands-on experience in RL training, scalable ML systems, agent environments, evaluation, inference or training infrastructure, or research spanning methods and systems.
Experience building and debugging model-training, post-training, agent, evaluation, or closely adjacent experimental loops.
Strong experimental judgment, including meaningful baselines, careful trace inspection, clear failure attribution, and willingness to change direction when evidence changes.
Strong Python skills and ability to build artifacts that other researchers and engineers can run, inspect, and extend.
Comfort owning ambiguous problems and keeping partners and teammates current as problems are resolved.

Nice To Haves

Evidence of exceptional work, such as publications, open-source systems, benchmarks, or shipped products.

Responsibilities

Develop reinforcement learning and post-training methods for tool-using, multi-agent, memory-augmented, and long-horizon systems.
Build agentic judges and credit-assignment methods that inspect complete trajectories, use tools, and provide useful learning signals beyond a single final score.
Make multi-agent and long-horizon environments train reliably at scale while preserving debuggable traces and comparable baselines.
Identify apparent gains caused by reward hacking, harness behavior, data leakage, or infrastructure artifacts, and redesign experiments so results are robust.
Develop reusable environments, harnesses, rewards, and evaluations.
Run experiments connecting algorithm choices to training and inference behavior.
Build open-source systems and papers from problems surfaced by application researchers and external users.
Work with customers, users, research partners, application teams, and teammates, keeping partners and teammates current while resolving ambiguous problems.