The opportunity
The Personalization team makes deciding what to play next easier and more enjoyable for every listener. From Blend to Discover Weekly, we’re behind some of Spotify’s most-loved features.
What you'll do
You'll build and improve the core agentic capabilities that power the agent: behind Talk to Spotify (memory, context management, multi-step tool use)
You'll design and calibrate evaluation frameworks (including LLM-as-judge): that accelerate our confidence in the agent's behavior, and increase our offline-to-online success
You'll work in a very dynamic space: the team prototypes, dogfoods, ships, learns, and refines in tight loops with real users, as our understanding of the problem and users' expectations of agentic products and Spotify evolve
What they're looking for
- You're excited by agentic experiences: building agents, evaluating agents, and the hard problems in between (context handling, multi-step reasoning, ambiguity at scale)
- You like getting your hands dirty: shipping quickly, testing ideas against real usage, and learning from the wild rather than over-indexing on offline evaluation
- You have 5+ years of production ML experience deploying highly impactful: products, or equivalent experience in other roles with a deep ML background
- You know how to evaluate ML systems rigorously: designing metrics, building eval pipelines, judge alignment, and can develop intuition through dogfooding and looking at user behavior
- You're comfortable debugging the messy interactions between models, tools, and system constraints like latency
