RAG: search and answers based on your documents.
Retrieval-Augmented Generation means the model pulls the relevant fragments from your documents before answering and answers from them, with the source shown. That removes the main objection to LLMs in work tasks: you can't tell where an answer came from or whether to trust it.
What it consists of
Preparing the corpus
Parsing documents, chunking, removing duplicates and outdated versions. The boring part — and quality depends on it more than on the choice of model.
Retrieval that finds the right thing
Vector search alone often misses on terminology and part numbers. A hybrid with keyword search plus reranking is usually required.
An answer with its source
The user sees which document and which passage the answer came from. Without that, people don't start trusting the system no matter how much you improve it.
Quality evaluation
A set of real questions with correct answers, showing whether a change made things better or worse. Otherwise improvements are made blind.
Why us
FAQ
What is RAG in plain words?
Before answering, the model searches your documents for relevant pieces and answers from them. Like an employee who checks the manual before replying instead of answering from memory.
Why RAG rather than fine-tuning?
Documents change. RAG updates with them: put in a new version of the manual and the system answers from it. Fine-tuning can't do that and costs more.
Will our documents leave our perimeter?
Only if you allow it. For sensitive corpora we use self-hosted models or a setup where a minimum leaves the perimeter.
How many documents make this worthwhile?
It's not about the count but about how much time people spend searching. If an answer is found in half a minute by ordinary search, RAG isn't needed.
Read also
RAG vs fine-tuning: what should a knowledge assistant use?
If an assistant must answer from changing documents, start by checking retrieval and sources. If the problem is consistent output format or behaviour, test instructions and then consider fine-tuning. RAG and fine-tuning can work together; the choice depends on the failure you need to fix.
How to evaluate a RAG system before production
Evaluate RAG with questions people already ask of your documents. For each question, record the correct source, an acceptable answer and when the system should refuse. Measure retrieval and the final answer separately. That tells you whether to fix data, search or the model.
How to update a RAG knowledge base without stopping the system
A demo lives on a static corpus, production on a moving one. Incremental updates instead of full reindexing, swapping the index without a restart, deleting documents, and which metrics reveal retrieval degradation before users report it.
Chunking documents for RAG: why chunks matter more than the model
A bad RAG answer almost always comes from chunking rather than from the model. How to split policies, contracts, scanned PDFs and support tickets, what breaks on a corpus of tens of thousands of long documents, and how to tell that chunking is the culprit.
Checklist: are your documents ready for RAG
PDF, 2 pages. No sign-up, no email — just the file.