Amazon Bedrock Knowledge Bases helps connect foundation models to an organization’s information for retrieval-augmented generation (RAG). Instead of relying only on what a model learned during training, an application retrieves relevant source material and uses it to help generate a grounded response.
How the workflow works
- Connect supported data sources.
- Prepare and split documents into chunks.
- Create embeddings and store them in a supported vector store.
- Retrieve relevant chunks for a user’s query.
- Pass the retrieved context to the model to generate a response.
Knowledge Bases vs. fine-tuning
Choose Knowledge Bases when answers should use current or private reference information and the model should rely on retrieved context. Fine-tuning is for adapting model behavior or patterns using training examples; it is not usually the first choice simply to make changing company documents available.
Exam decision clues
Look for phrases such as “answer questions from internal documents,” “ground responses in company data,” or “retrieve relevant sources.” Those clues point toward RAG and Knowledge Bases. A vector database stores and searches embeddings; embeddings represent semantic features of data.
Limitations
RAG can improve grounding but does not eliminate hallucinations. Quality depends on source documents, chunking, retrieval settings, permissions, and the model’s use of retrieved context.
Related topics
Review RAG vs. fine-tuning, tokens, embeddings, and vector databases, and Amazon Q Business vs. Amazon Q Developer.
Prepare for the AWS Certified AI Practitioner (AIF-C01)
Test your understanding of AWS AI services and responsible AI concepts with original practice questions and detailed explanations.
Try the free AIF-C01 diagnostic or explore the 390-question premium practice package.