GenRec: An LLM-Backed Recommendation Ranker at Netflix

2026-08-10

AI Summary

This work addresses the limitations of traditional recommendation rankers, which rely heavily on manual feature engineering and struggle to capture the complex semantic relationships among users, items, and contextual signals. To overcome this, the authors propose a two-stage ranking framework grounded in large language models (LLMs): first, an internal base LLM is adapted to Netflix data, followed by post-training using recommendation-specific tokens and reward signals. This approach reformulates user history and contextual information into natural language representations and employs a prefill-only inference architecture. The methodology shifts the paradigm from feature engineering to context engineering and from bespoke architectures to shared LLM foundations. Large-scale A/B experiments demonstrate that, with only minimal labeled data, the model significantly outperforms the existing production system across both offline and online evaluation metrics.

Abstract

Large language models (LLMs) are reshaping recommender systems by enabling richer modeling of users, content, and context directly in natural language. At Netflix, we are exploring this direction through GenRec, an LLM-backed recommendation ranker built on top of an in-house foundational LLM. GenRec follows a two-phase framework: Phase 1 adapts an open-source LLM to Netflix data, developing deep understanding of the catalog and member behavior while balancing capabilities such as content understanding and instruction following. Phase 2 post-trains this foundation model with recommendation-ranking specific data, labels, and reward signals, aiming to align the ranker with business requirements and long-term member satisfaction. This paper focuses on Phase 2 and the transition from a traditional discriminative ranker with thousands of engineered features to an LLM-backed ranker driven by verbalized user histories and context. We describe our design for input verbalization and context engineering, post-training data construction, reward integration, model architecture, and a cost-constrained serving design based on a prefill-only inference approach. We report results from a large-scale A/B test comparing GenRec against the current production ranker model, where we show that a GenRec model trained with substantially fewer Phase-2 labeled training examples and input signals can achieve statistically significant gains in offline and online metrics. We discuss how LLM-backed recommenders could shift the recommendation paradigm: from feature engineering to context engineering, and from bespoke architectures to shared foundation backbones. We also outline practical lessons for serving such systems under real-world resource constraints.

Problem

• LLM-backed recommender
• recommendation ranking
• context engineering
• natural language modeling
• feature-to-context shift

Innovation

• LLM-backed recommender
• context engineering
• verbalization
• post-training
• prefill-only inference