Applied Research - Agentic RL
$150k - $300k • Remote • San Francisco • FullTime
Posted 1d ago
Job Location
San Francisco
Tech Stack
Remote Work Policy
Fully remote
Employment Type
FullTime
Categories
AI Research Engineer
About the job
Prime Intellect is developing an open superintelligence stack, providing infrastructure that AI labs typically build internally to ambitious AI teams. Their platform, Lab, integrates compute, environments, evaluations, secure sandboxes, high-performance training, and deployment for post-training at scale, covering SFT, RL, tool use, agent workflows, and continuous model improvement. They are focused on building open-source frontier AI models trained for long-horizon tasks and the platform their research team uses. The company has secured significant funding from prominent investors and AI leaders, and is seeking individuals passionate about building at the intersection of frontier research, infrastructure, and go-to-market strategy for a nascent category.
Responsibilities
- Build agentic judges and credit-assignment methods for inspecting trajectories, using tools, and assigning learning signals beyond a single score.
- Ensure multi-agent and long-horizon environments train reliably at scale while maintaining debuggable traces and comparable baselines.
- Identify and address issues like reward hacking, harness behavior, data leakage, or infrastructure artifacts by redesigning the learning loop.
- Develop RL and post-training methods for systems involving tool use, multi-agent coordination, memory augmentation, and long horizons.
- Create reusable environments, harnesses, rewards, and evaluation frameworks.
- Conduct experiments linking algorithm choices to training and inference behavior.
- Contribute to open-source systems and research papers based on problems identified by application researchers and external users.
Requirements
- Hands-on experience with RL training, scalable ML systems, agent environments, evaluation, or inference/training infrastructure.
- Experience building and debugging model-training, post-training, agent, evaluation, or related experimental loops.
- Strong experimental judgment, including establishing meaningful baselines, inspecting traces, attributing failures, and adapting approaches based on evidence.
- Proficiency in Python and the ability to create runnable, inspectable, and extensible artifacts for other researchers and engineers.
- Comfort managing ambiguous problems and keeping stakeholders informed during resolution.
Benefits
- Cash Compensation Range of $150-300k
- Equity incentives
- Flexible Work (San Francisco, New York, London, or remote)
- Visa Sponsorship & relocation support
- Professional Development budget
- Team Off-sites & conference attendance