Source
GenRec: Towards LLM-Native Recommendation at Netflix
Netflix developed GenRec, an LLM-backed recommendation ranker that transforms user history and item metadata into natural-language prompts. The system uses a two-phase training process to produce personalized content rankings.
Why it matters
By reducing reliance on manual feature engineering and large labeled datasets, Netflix can improve recommendation accuracy more efficiently. This results in more precise content discovery for members based on their actual viewing habits.
The details
- GenRec outperformed a production ranker in A/B tests across 10% of traffic.
- The system requires 10-40x fewer Phase-2 labeled examples than the production baseline.
- It uses prefill-only inference via vLLM to minimize serving costs.
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.