Retrieval-augmented generation: grounding language models in your own documents and data.
The pipeline end to end — index, retrieve, assemble, generate — and where each stage fails.
Splitting a corpus so retrieved passages are self-contained, and keeping the index current.
Getting the right passage into the top few, and measuring whether you did.
Making answers traceable to sources, and making the system say "I do not know".
Building the evaluation that tells you which stage to fix, and running it as a regression test.
Answering questions about numbers, where retrieving prose is the wrong tool.
Multi-turn question answering: rewriting, history, memory and the state that makes it coherent.
Making sure an assistant never surfaces something the asker may not see.
Handling PDFs, scans, images and documents far larger than any context window.
35 in this topic
A map of a complete retrieval-augmented generation system, from ingestion to answer, and what each component is responsible for.
RAG 1 min read 14 Dec 2025 For people
Turning PDFs, slides, HTML and scans into clean, structured text — the unglamorous step that decides RAG quality.
RAG 2 min read 13 Dec 2025 For people
How to choose chunk size and overlap for retrieval by testing against real questions rather than guessing.
RAG 2 min read 12 Dec 2025 For people
Combining keyword and vector search so RAG finds both exact terms and paraphrased meaning.
RAG 2 min read 11 Dec 2025 For people
Improving retrieval by rewriting user questions: expansion, decomposition, hypothetical answers and conversational context.
RAG 1 min read 10 Dec 2025 For people
Using document metadata — dates, sources, products, permissions — to narrow retrieval and get more relevant, safer results.
RAG 2 min read 9 Dec 2025 For people
How to make RAG answers cite their sources reliably, and how to check that citations really support the claims.
RAG 2 min read 8 Dec 2025 For people
How to measure a RAG system properly: retrieval quality, answer correctness, faithfulness and the test set that ties them together.
RAG 2 min read 7 Dec 2025 For people
Keeping answers tied to retrieved sources: prompting, checking and measuring whether every claim is supported.
RAG 2 min read 6 Dec 2025 For people
Incremental updates, deletions, versioning and monitoring so a RAG system never answers from outdated documents.
RAG 1 min read 5 Dec 2025 For people
Making sure RAG only retrieves content each user is allowed to see, and avoiding leaks through prompts, caches and logs.
RAG 2 min read 4 Dec 2025 For people
Answering questions over databases and spreadsheets with text-to-SQL and table retrieval instead of text chunks.
RAG 2 min read 3 Dec 2025 For people
When a single retrieval step isn't enough: letting an agent decide what to search, when, and how often.
RAG 1 min read 2 Dec 2025 For people
Using knowledge graphs alongside vector search to answer questions about relationships and whole collections.
RAG 2 min read 1 Dec 2025 For people
With context windows now very large, when should you just paste in the documents, and when is retrieval still better?
RAG 2 min read 30 Nov 2025 For people
Why chunks lose meaning when cut from their documents, and how prepending context improves retrieval accuracy.
RAG 2 min read 29 Nov 2025 For people
The recurring ways RAG systems fail — from missed retrieval to stale content — and how to diagnose each one.
RAG 1 min read 28 Nov 2025 For people
The factors that matter when picking an embedding model for retrieval: language, domain, dimensions, input length, cost and licence.
RAG 2 min read 27 Nov 2025 For people
How to structure the prompt that turns retrieved passages into a grounded, cited answer.
RAG 1 min read 26 Nov 2025 For people
Making RAG work in multi-turn chat: rewriting follow-ups, carrying context and avoiding stale retrieval.
RAG 1 min read 25 Nov 2025 For people