RAG Pipelines That Actually Work: Lessons From Shipping Retrieval Systems
Chunking strategy, hybrid search, reranking, and evaluation — the unglamorous decisions that separate RAG demos from systems users actually trust.
Topic
#architecture (2)
Everything I've written on this topic.
[02/02]
Building AI Agents That Actually Ship: A Practical Architecture
Most agent demos die outside the playground. Here's the production architecture I use — tool schemas, guardrails, memory, and evaluation — that survives real users.