Topic
RAG and Knowledge AI
Retrieval-augmented generation is how most enterprises get an AI system that answers from their own documents without retraining a model. The model supplies the language, your data supplies the facts, and every claim can carry a citation back to its source.
The quality of a RAG system is decided less by the model and more by the parts around it: how documents are chunked, whether search is hybrid, whether results are reranked, and how rigorously the answers are evaluated for groundedness rather than fluency. These articles cover what actually moved the numbers.
What we would tell you on a call
Retrieval quality sets the ceiling
No amount of prompt work rescues a system that hands the model the wrong five passages. Most RAG projects that disappoint are retrieval projects that were never measured as retrieval projects.
Filter before you search
Metadata filtering ahead of the vector search, by jurisdiction, entity, date or document type, is repeatedly the highest-leverage change available. It shrinks the candidate set to the documents that could possibly be right before similarity is asked to do anything.
Citations are not a feature, they are the architecture
If every generated claim links to the passage it came from, a reviewer can check the system in seconds and a wrong answer is visible rather than plausible. Designing for that from the start changes how the retrieval and the generation are built.
Freshness is an ongoing cost
A knowledge system is only as current as its last re-index. Someone has to own the ingestion pipeline, the re-index schedule and what happens when a source document is withdrawn.
Everything we have written on this
RAG Evaluation Metrics That Matter
RAGAS scores were a good starting point in 2023. The teams that avoided painful incidents added domain-specific metrics for faithfulness, context precision, and hallucination rate.
RAG Patterns for Legal, Banking, and SaaS
Between 2022 and 2024 we implemented RAG systems in legal, banking, and B2B SaaS. The architectures looked different, but the success patterns were surprisingly consistent.
Proof in production
Key terms
- Retrieval-augmented generation (RAG)
- A pattern where the system searches your own content for passages relevant to a question, then gives those passages to a language model to answer from. The model supplies the language, your data supplies the facts.RAG is how most enterprises get an AI system that answers from their documents without retraining a model.
- Chunking
- Splitting source documents into passages small enough to retrieve precisely and large enough to stay meaningful on their own.Chunking strategy affects answer quality more than the choice of model in most RAG builds.
- Reranking
- A second pass that reorders retrieved passages by relevance to the specific question, usually with a model built for scoring rather than generation.Reranking is often the cheapest single improvement available to a RAG system that returns roughly right but not quite right answers.
- Citation architecture
- Designing a system so that each statement it produces carries a link to the source document, section, or record it came from.In regulated work, an uncited answer is unusable regardless of whether it happens to be correct.
The service behind this
RAG Systems & Knowledge AI
AI that answers from your data, with citations, not guesses.
