Best of Netflix — July 2025

1. 1
   Article

   Netflix TechBlog·41w

   Netflix Tudum Architecture: from CQRS with Kafka to CQRS with RAW Hollow

   Netflix migrated their Tudum fan site architecture from a CQRS pattern using Kafka and traditional caching to RAW Hollow, an in-memory compressed object database. The original architecture suffered from eventual consistency delays, taking minutes for content changes to appear. RAW Hollow eliminated the need for separate databases and Kafka infrastructure by storing the entire dataset in memory across application processes, reducing homepage construction time from 1.4 seconds to 0.4 seconds and enabling real-time content previews.

   91
2. 2
   Article

   Netflix TechBlog·40w

   Behind the Streams: Three Years Of Live at Netflix. Part 1.

   Netflix built a comprehensive live streaming architecture over three years, handling events from comedy specials to NFL games and boxing matches. The system leverages dedicated broadcast facilities, cloud-based transcoding pipelines using AWS MediaConnect and MediaLive, the Open Connect CDN for global delivery, and HTTPS-based streaming with AVC/HEVC codecs. Key learnings include the importance of extensive testing, regular practice events, viewership prediction, graceful degradation strategies, and comprehensive contingency planning with dedicated launch rooms and game day exercises.

   13
3. 3
   Video

   Coding with Lewis·42w

   3 Insane Algorithms Netflix Uses to Scan BILLIONS of Frames

   Netflix uses three sophisticated computer vision algorithms to analyze billions of video frames: match cut transitions that automatically find visually similar shots for seamless editing, video search capabilities that convert text queries into mathematical embeddings to find specific scenes, and scene detection that combines screenplay alignment with multimodal analysis of video and audio tracks. These systems leverage instance segmentation, optical flow, and bidirectional neural networks to automate video editing tasks that would otherwise require thousands of manual hours.

   14