A knowledge base is a centralized collection of information that an AI system can store, organize and use to answer questions or solve tasks. It acts as the AI's memory of facts, documents and data.
In simple terms
The AI's organised store of facts and documents that it looks into to answer questions.
How it works
- 1Collect: gather information from various sources.
- 2Process: clean, chunk and convert data into embeddings.
- 3Store: store embeddings and metadata in a vector database.
- 4Retrieve: search the knowledge base to find the most relevant information.
- 5Use & answer: pass the retrieved information to the model to generate an accurate response.
Key points
- Sources include documents (PDF, DOCX), web pages and databases.
- It is continuously updated, so answers stay current.
- Sources can be traced and verified, improving trust.
- One knowledge base can power many applications.
Why it matters
A well-built knowledge base grounds an AI in reliable, up-to-date information. It's what turns a general model into an expert on your company's or domain's specific content.
Frequently asked questions
- How is a knowledge base different from training data?
- Training data shapes the model's general ability; a knowledge base is external, updatable information retrieved at query time.
- What types exist?
- Document KBs, web KBs, structured (database) KBs and code KBs.