Skip to content

RAG in Production: Operations & Case Studies

Theory is great, but production is where the rubber meets the road. This section curates deep-dive engineering blogs, whitepapers, and talks from companies running Retrieval-Augmented Generation at scale.

Criteria: We only include resources that discuss architecture, latency, evaluation, or scaling challenges. No marketing fluff.


High-Scale Consumer Apps

Perplexity.ai

  • Perplexity's Online LLM Inference
  • Serving 70B models with low latency for real-time search — key techniques: speculative decoding, aggressive caching, and search index optimization.

Notion AI

  • Notion AI
  • How Notion integrated RAG into a collaborative workspace for millions of non-technical users — agents, search, and knowledge management in one product surface.

Discord

  • How Discord Scaled Vector Search
  • A masterclass in ANN retrieval at trillion-message scale — Rust-based microservices with ScyllaDB for metadata, used in Clyde and search.

Enterprise & B2B

Stripe

  • Stripe Radar & ML Infrastructure
  • Using embeddings for fraud detection (retrieval-based classification).
  • Real-time feature extraction and low-latency vector lookups at payment scale.

Airbnb

  • Building Airbnb's AI Search
  • The evolution of search ranking with embeddings — hard negative mining and custom listing embeddings for domain-specific recall.

Lessons from the Trenches (Engineering Blogs)


Must-Watch Talks


(back to main resource)