Member of Technical Staff (Model Behavior Architect)

About the Role We're looking for a Model Behavior Architect to help build Perplexity's AI products and evaluations. You'll sit within our AI team and collaborate closely with research and product teams, designing prompt and context engineering strategies to deliver high quality user experiences across multiple domains and models. This role is equal parts craft and science. You'll develop a deep understanding of our answer engine by pressure-testing model capabilities and working across our AI infrastructure (including system and tool prompts, skills, and evaluations) to create a stellar product experience for our users. You'll serve as a go-to expert on prompting, model quality, and behavioral consistency across new product features and model releases. Key Responsibilities - Context Engineering: Design, test, and optimize context strategies and system prompts that shape answer engine behavior across products, features, and use cases. - Evaluation Systems: Build automated and semi-automated evaluation pipelines that measure model quality, catch regressions, and scale across product surfaces. - Model Launch Support: Partner with research and engineering to validate model behavior before and during rollouts, ensuring smooth transitions with no degradation. - Research & Analysis: Identify inconsistencies and failure modes in model outputs through well-designed research projects — for both internal and production-facing systems. - Cross-functional Collaboration: Work closely with design, product, and research teams to translate product goals into concrete model behavior requirements. - Knowledge Sharing: Help engineers across teams build intuition for prompt design, context engineering, and evaluation best practices. - Staying Current: Track the latest alignment, evaluation, and prompting techniques from industry and academia, and bring the best ideas back to the team. What We're Looking For Required - Experience designing evaluations, benchmarks, or metrics for AI systems. - Strong written and verbal communication skills, particularly in explaining complex concepts to diverse stakeholders. - Ability to manage multiple concurrent projects in a fast-moving environment. - Strong experience with Perplexity or other frontier AI models in production settings. - Demonstrated experience with Python — you'll prototype, debug, automate, and build systems at scale. - 3+ years of experience working with LLMs in a product or research setting. Preferred - Experience with A/B testing or experimentation frameworks. - Track record of improving AI system performance through systematic evaluation and iteration. This Role May Be a Great Fit for You if You - Get excited about edge cases in model behavior and love digging into how an answer could be better. - Enjoy turning qualitative "this feels off" intuitions into quantitative metrics and systematic fixes. - Want to work at the intersection of research and product, where your work ships to real users same-day. - Are comfortable with ambiguity and can define what "good" looks like for novel AI features. - Have a hacker spirit — you'd rather build a quick prototype to test a hypothesis than debate it in a doc. - Care deeply about making AI more reliable and useful for our users.

Experience designing evaluations, benchmarks, or metrics for AI systems. Strong written and verbal communication skills, particularly in explaining complex concepts to diverse stakeholders. Ability to manage multiple concurrent projects in a fast-moving environment. Strong experience with Perplexity or other frontier AI models in production settings. Demonstrated experience with Python — you'll prototype, debug, automate, and build systems at scale. 3+ years of experience working with LLMs in a product or research setting.

Context Engineering: Design, test, and optimize context strategies and system prompts that shape answer engine behavior across products, features, and use cases. Evaluation Systems: Build automated and semi-automated evaluation pipelines that measure model quality, catch regressions, and scale across product surfaces. Model Launch Support: Partner with research and engineering to validate model behavior before and during rollouts, ensuring smooth transitions with no degradation. Research & Analysis: Identify inconsistencies and failure modes in model outputs through well-designed research projects — for both internal and production-facing systems. Cross-functional Collaboration: Work closely with design, product, and research teams to translate product goals into concrete model behavior requirements. Knowledge Sharing: Help engineers across teams build intuition for prompt design, context engineering, and evaluation best practices. Staying Current: Track the latest alignment, evaluation, and prompting techniques from industry and academia, and bring the best ideas back to the team.