Ship it

You built direct chat, native-tool RAG, and a custom pipeline in one Next.js app. Before you close this tab, let us lock in the decision framework and talk about what changes when you go to production.

production-notes.md

text

Swap MemoryVectorStore for a real store
  Qdrant, pgvector, Chroma, Pinecone, Weaviate all have LangChain adapters.
  The interface barely changes, the operability story changes a lot.

Persist documents
  Scraping on every request is wasteful. Store documents and re-embed on write, not on read.

Rerank before grounding
  Add a reranker like Cohere Rerank between similaritySearch and the prompt for better faithfulness on ambiguous queries.

Evals
  Build a small eval set with questions and expected snippets. Rerun your benchmark after every prompt or chunking change.

Structured logging
  Log request id, route, retrieval hits, and first-token latency. You cannot improve what you cannot see.

These are the steps most teams hit right after finishing a workshop like this one. Pick the ones that match your next real project.

Checkpoint: Final checkpoint

Loading practice…