Retrieval (RAG)

Retrieval examples enact the layers of a RAG pipeline. Sources under examples/retrieval/.

Name	Scenario	Source
`retrieval_text_splitters`	Compare `RecursiveCharSplitter`, `MarkdownSplitter`, `SentenceSplitter` on the same Markdown blog post — see how each carves it.	src
`retrieval_rag_pipeline`	End-to-end RAG — split, embed, store, retrieve, answer with the LLM. The canonical flow.	src
`retrieval_indexing_rag`	Docs-site re-indexer — round 1 indexes 3 docs, edit one, round 2 only re-embeds the changed doc.	src
`retrieval_reranking`	Vector search returns top-10; cross-encoder (LLM-judge) reranks to top-3 for the prompt.	src
`retrieval_caching_retriever`	Chat session asks the same question twice — second call returns from cache (latency drops to ~0).	src

How to run

COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
  cargo run -p cognis-examples --example retrieval_rag_pipeline

First time touching RAG? retrieval_rag_pipeline is the canonical flow.
Re-indexing a corpus? retrieval_indexing_rag — incremental updates with a RecordManager.
Improving retrieval quality? retrieval_reranking adds a cross-encoder pass.

The user guide for every layer.

A worked RAG over a Rust codebase.

Why reranking matters.