Glossary
Chunking
Splitting documents into passages small enough to embed and retrieve precisely.
Chunking is the process of breaking long documents into smaller passages before they are embedded and indexed. Chunks need to be small enough that each captures a focused idea, but large enough to retain context. Overlapping chunks slightly preserves meaning that would otherwise be cut across a boundary.
CertifChat chunks content at roughly 512 tokens with a 10–20% overlap, which balances retrieval precision against context. Well-sized chunks mean the model receives tight, relevant passages, which improves answer quality and keeps token usage (and cost) under control.
See how this works in practice across CertifChat's features.
Related terms
- EmbeddingsNumeric vectors that capture the meaning of text so similar passages sit close together.
- Retrieval-Augmented Generation (RAG)Grounding an LLM's answers in retrieved source documents instead of its training data alone.
- Semantic SearchSearching by meaning using embeddings, often combined with keyword search.
- Knowledge BaseThe collection of content an assistant draws on to answer questions.
Put your knowledge to work today
Connect your sources, drop one snippet on your site, and start deflecting tickets. We are opening CertifChat to a first group soon.