Spotify
Global music and audio streaming platform. Search challenge: the boundary between search and recommendations is blurry — users say “search” but often mean “recommend.”
Search Context
Exploratory queries like “new releases for me” or “similar artists to X” don’t fit traditional search — they require personalized retrieval that lexical or semantic matching alone can’t provide.
Key Engineering Work
Agentic LLM Router
- LLM router interprets complex query intent and routes to the appropriate downstream service: search API vs. recommendation API
- Post-training adaptation: router fine-tuned for production scale
- Specialized sub-agents handle specific exploratory intents
Results
| Use Case | Improvement |
|---|---|
| Similar artists discovery | +115% |
| New music releases | +91% |
| Broad music searches | +25% |
| Broad podcast searches | +15% |
Key Insight
Users express exploratory intent through a search interface. An agentic LLM router can identify whether the intent is truly a search or a recommendation, then delegate accordingly.
Articles
- You Say Search I Say Recs - Spotify Agentic Query Understanding — LLM routing architecture, results