AI Engineer
← All speakers

Bio, Work & Ideas

Yves Raimond

Conference affiliation: SVP/GM, AI & Personalization · Spotify · 2026

Yves Raimond was identified in his 2026 AI Engineer conference biography as Spotify’s SVP and GM of AI & Personalization, a senior leader for the company’s AI and personalization work. In his recorded presentation, he described earlier work on personalization for Google Search and at Netflix.

In “Teaching LLMs to Speak Spotify,” presented with Jacqueline Wood, Raimond explained Spotify’s shift from curated playlists and ranked recommendations toward generative personalization: experiences users can shape through natural language. He demonstrated a steerable DJ, prompted playlists, an editable taste profile that lets listeners correct inferred preferences and express new interests, and personal podcasts that generate tailored content. He described the shared Large Taste Model behind these experiences as combining prediction with reasoning about a listener’s history and context, allowing users to shape recommendations and generated experiences in real time. At the time of the talk, he reported that roughly one in four US Premium subscribers interacted with the system daily.

1 conference talk

▶ Watch ↗

AI Engineer World's Fair 202619:40

Teaching LLMs to Speak Spotify — Yves Raimond & Jacqueline Wood, Spotify

Read the full talk →

Key ideas

Scroll to read ↓

Yves Raimond and Jacqueline Wood explain how Spotify connects language models to its catalog, preserves their language abilities during training, and evaluates recommendations that users can steer and question.

  • Natural-language controls let listeners correct the interpretation of listening history and express interests that their past behavior does not yet capture.
    5:37 ↗
  • Semantic IDs connect catalog entities to the LLM’s vocabulary, allowing one response to contain both a recommendation identifier and a natural-language explanation.
    8:28 ↗
  • NEO grounds new catalog embeddings against a frozen backbone before multitask tuning. Spotify’s ablations show why task performance and retained language capabilities need separate attention.
    10:26 ↗
  • Decoding choices serve different goals: constraints can restrict eligible content, while Spotify favored beam search over top-p sampling for recommendation accuracy despite added latency.
    14:27 ↗
  • Evaluation must cover intent, fit and explanation accuracy. Grounding LLM judges in listener profiles and past behavior makes those judgments better informed, especially for ambiguous requests.
    16:37 ↗

References