AI BriefingNetflix Tech news
The latest AI and developer news from Netflix Tech, with short summaries.
Visit website →Modeling Device Capabilities for Analysis
Netflix has built a data model that finely manages feature support across different devices.
GenRec: Toward Netflix's LLM-Native Recommendations
Netflix has presented a direction for transitioning its complex existing recommendation system into an LLM-native structure.
GenRec: Toward Netflix's LLM-Native Recommendation System
Netflix proposes GenRec, a model that moves away from its existing complex feature-engineering-centered recommendation system and puts LLMs at its core.
Netflix's In-House LLM Serving Platform
Netflix built an efficient LLM serving system by integrating vLLM and Triton into its existing ML infrastructure.
PickNetflix's In-House LLM Serving Infrastructure
Instead of using hosted APIs, Netflix built its own LLM serving infrastructure to directly operate the entire stack from model deployment to inference.
Building Large-Scale Service Topology: Architecture, Challenges, and Lessons
Netflix built a real-time service dependency map to manage its complex microservices environment.
GenPage: Toward Netflix's End-to-End Generative Home Screen
Netflix is introducing GenPage technology, which sequentially generates rows and entities to construct personalized home screens.
Towards More Controllable AI Video Editing: Exploring Netflix's Early Research
Netflix unveiled two research models designed to address the limitations of existing generative AI—unintended alterations and physical discontinuity.
How Netflix Simplified Batch Computing with Kueue
Netflix modernized its batch workload management system by adopting Kubernetes-based Kueue in place of its self-built CMB.
Data Canary: How Netflix Validates Catalog Metadata
Netflix built a real-time traffic-based 'Data Canary' system to validate the integrity of data pipelines, not just code deployments.
Data Projects: Managing Data Assets at Netflix Scale
To address the complexity of managing data assets, Netflix introduced Data Projects, a unified management system organized around projects rather than individual asset units.
Predicting Content Launch Risk: Transforming Release Planning with Data-Driven Insights
Netflix uses production data and predictive models to reduce uncertainty in content release schedules and manage risk.
The Evolution of Netflix's Cassandra Data Migration Technology
Netflix has redesigned its data migration approach from Cassandra to Iceberg to overcome the limitations of the existing Casspactor.
Thinking Fast & Slow for Personalized Notification Systems
Netflix built a hierarchical notification system that separates a 'Slow' policy responsible for strategic planning from a 'Fast' policy responsible for real-time execution, in order to jointly optimize short-term engagement and long-term user experience.
Human-Augmenting Agent Workflow for Causal Inference
Netflix has unveiled an agent workflow that assists expert judgment to improve the accuracy of causal inference.
VMAF v1: Good Enough Isn't Enough
Netflix has open-sourced VMAF v1, which overcomes the limitations of the existing VMAF and improves detection of compression artifacts and responsiveness to viewing environments.
Dynamic Splitting of Wide Partitions in Cassandra for Time-Series Workloads
To solve the wide partition problem that occurs in Cassandra when processing time-series data, Netflix introduced a technique to dynamically split partitions.
Netflix's High-Throughput Graph Abstraction: Part I
Netflix built an OLTP-optimized Graph Abstraction that handles 10 million operations per second.
Beyond Silos to Service Topology: Why Netflix Built a Real-Time Service Map
Netflix built a real-time service topology map to understand dependencies in its complex distributed infrastructure and speed up incident response.
Extending ArchUnit with Nebula ArchRules
Netflix introduced a way to share ArchUnit rules across tens of thousands of Java repos using Nebula ArchRules.