About the role
Perplexity is seeking an experienced Machine Learning Engineer to advance how AI systems search, reason, and work together to solve complex problems. Our work spans search and retrieval, LLM post-training, multi-agent training, and the harnesses that make these systems effective.
We control the full stack: the models, the agent harnesses, and the search infrastructure underneath. That gives us the freedom to develop new approaches across all three training models to use search more effectively, designing tools and execution environments around learned behavior, and improving retrieval to support how agents actually work.
Responsibilities
- Push search and agent quality forward through improvements to models, training data, tools, and system design.
- Develop LLM post-training methods, including reinforcement learning, to improve reasoning, search, tool use, and task completion.
- Train and evaluate multi-agent systems, exploring how agents divide work, share information, and coordinate effectively.
- Design and build agent harnesses around the tools, context management, execution environments, and orchestration that support reliable work over many steps.
- Improve retrieval and ranking models and the search interfaces agents use to find and assess information.
- Build datasets, reward signals, and evaluations that expose meaningful failures and guide improvements.
- Own experiments end to end, from a clear hypothesis to scalable training, deployment, and measurable gains in quality, latency, and cost.
- Collaborate with AI, Search, Infrastructure, Data, and Product teams to bring new capabilities into production.
Qualifications
- A strong track record of building and shipping ML systems, with deep experience in one or more of LLM post-training, reinforcement learning, search and retrieval, or agent systems.
- Strong software engineering skills and the ability to work across model training, experimentation infrastructure, and production systems.
- Experience designing rigorous evaluations, diagnosing failures, and translating experimental results into practical improvements.
- Comfort with open-ended problems that require both research judgment and hands-on engineering.
- A strong sense of ownership, curiosity, and the drive to carry an idea through to a working system.
Nice to have
- Experience with training models to use tools or complete tasks over many steps.
- Multi-agent training, coordination, or evaluation.
- Building agent harnesses, distributed training systems, or scalable inference infrastructure.
- Large-scale retrieval, ranking, or recommendation systems.