Applied Research - Continual Learning

Prime Intellect•San Francisco, CA

1d•$150,000 - $300,000

About The Position

Prime Intellect is hiring an applied researcher to build continual learning systems that transform agent trajectories, corrections, and failures into new tasks and model updates. The role compares streaming reinforcement learning, batched post-training, and distillation, and determines whether improvements should be incorporated into model weights or represented through persistent memory, skills, prompts, or subagent state. Experiments will track skill acquisition, transfer, forgetting, and loss of plasticity across long sequences of tasks. Applied Research at Prime Intellect turns workflows into tasks, environments, evaluations, and reward signals; runs post-training experiments against real baselines; and deploys effective results. Researchers investigate whether observed gains come from the model, data, harness, inference, or infrastructure. Reusable work may become shared research, open-source systems, or product capabilities. Applied Research roles are externally facing, and most researchers work directly with customers, users, or research partners while maintaining a substantive research agenda. The researcher will control update cadence, policy lag, task drift, and exploration; evaluate checkpoints on earlier skills, unseen tasks, and general capabilities; and build reproducible task-generation and training pipelines. The role includes working directly with external collaborators whose deployments drive the research. Its scope also includes methods for selective consolidation, replay, adaptation, memory, and online or offline RL; evaluations and promotion gates for agents that continue changing in production; and field experiments, benchmarks, and open research that make continual improvement measurable and transferable.

Requirements

Relevant experience in continual or online learning, RL, meta-learning, memory systems, post-training data loops, adaptive production ML, or evaluation under changing distributions.
Hands-on 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 other researchers and engineers can run, inspect, and extend.
Comfort owning ambiguous problems and keeping partners and teammates current while resolving them.

Nice To Haves

Publications, open-source systems, benchmarks, or shipped products demonstrating exceptional work.
Ability to show a difficult project built, what failed, and how evidence changed the approach.

Responsibilities

Build continual learning systems that convert agent trajectories, corrections, and failures into new tasks and model updates.
Compare streaming RL, batched post-training, and distillation, and determine where adaptation should reside: model weights, harness state, memory, tools, or data.
Track skill acquisition, transfer, forgetting, and loss of plasticity over long sequences of tasks.
Control update cadence, policy lag, task drift, and exploration, and evaluate checkpoints on earlier skills, unseen tasks, and general capabilities.
Build task-generation and training pipelines with reproducible data and harness versions so regressions can be traced and rolled back.
Work directly with external collaborators whose deployments drive the research.
Select high-value production traces and interventions, convert them into training signals, and establish regression gates before deployment.
Build evaluations that distinguish retained learning and transfer from memorization, context accumulation, or measurement drift.
Develop methods for selective consolidation, replay, adaptation, memory, and online or offline RL.
Develop evaluations for retention, transfer, regression, and rollback as tasks and distributions change.
Establish promotion gates and human oversight for agents that continue to change in production.
Conduct field experiments and produce benchmarks and open research that make continual improvement measurable and transferable.