// Track

RAG in Production

Retrieval Systems That Show Their Receipts

Build retrieval-augmented generation that holds up in production — gated corpora, structure-aware chunking, vector indexing with a re-embed lifecycle, tuned retrieval, cited answers that refuse when retrieval is thin, golden-set evals, and the ops layer: cost, latency, injection defense, and access control.

Recommended: Complete Context Engineering & CLAUDE.md Architecture first

10 lessons~84 min total

Lessons are shown in recommended order. Complete them in sequence for the best experience — or jump to any lesson.

Lesson 576·7 min read

RAG Is a Production System, Not a Demo

RAG has a failure taxonomy, SLOs, and an owner — treat it like any other production service or it will lie to users at scale.

Lesson 577·7 min read

Corpus Design & Leak Gates: What Goes In Decides What Comes Out

The corpus is not a directory of documents — it is a curated, gated, access-controlled knowledge base where every source has an owner, a trust level, and a freshness cadence.

Lesson 578·7 min read

Chunking: The Decision That Haunts Every Query

Chunk size is the most consequential decision in a RAG pipeline — it determines retrieval precision and generation context quality, and changing it requires re-embedding everything.

Lesson 579·7 min read

Embeddings & the Index: The Semantic Layer

Embeddings are semantic coordinates — choose your model carefully because changing it means re-embedding everything, and choose pgvector by default because boring infrastructure is production infrastructure.

Lesson 580·8 min read

Retrieval Quality: Top-K Is Not a Strategy

Returning the top-K chunks by cosine similarity is a starting point, not an answer — learn how thresholds, reranking, and query rewriting turn approximate recall into precise, trustworthy context.

Lesson 581·9 min read

Grounded Generation: Answers That Show Receipts

A RAG answer without citations is an assertion without evidence — learn how to wire structured output schemas, parse defensively, and treat every citation as a verified receipt rather than a model suggestion.

Lesson 582·9 min read

The Re-Embed Lifecycle: Your Index Is Always Stale

The vector index is a cache of your corpus — and like every cache, it needs invalidation. Checksum-based incremental re-embedding, artifact freshness metrics, and the lesson that redaction completes at re-embed, not at merge.

Lesson 583·9 min read

Evaluating RAG: Faithfulness, Recall, and the Golden Set

A RAG system you cannot measure is a liability you are running blind — build the golden set, measure retrieval and generation separately, and gate every corpus change on regression scores.

Lesson 584·10 min read

RAG Operations: Cost, Latency, Injection & Access Control

Operating a RAG system in production means owning cost, latency, security, and access control — four concerns that interact with every stage of the pipeline and cannot be bolted on after launch.

Lesson 585·11 min read

Capstone: A RAG Service With Receipts

Assemble the complete RAG service: gated ingest, structure-aware chunking, pgvector index with checksums, threshold-and-rerank retrieval, cited generation with refusal, eval gate, and ACL — every component from every lesson, wired into one system.