I'm building a conversational agent with AWS Bedrock and its agents. I've linked a knowledge base to the agent to retrieve information based on user questions.
When testing its functionality, I see in the traces that the knowledge base always returns 5 chunks of information. However, I don't need that many - I only want the most relevant chunk.
How can I do this? What configuration do I need to modify?
I'm building a conversational agent with AWS Bedrock and its agents. I've linked a knowledge base to the agent to retrieve information based on user questions.
When testing its functionality, I see in the traces that the knowledge base always returns 5 chunks of information. However, I don't need that many - I only want the most relevant chunk.
How can I do this? What configuration do I need to modify?
Amazon Bedrock returns up to five results in the response by default. You can change the parameter numberOfResults in the vectorSearchConfiguration (if using python). By example:
response = bedrock.update_knowledge_base(
knowledgeBaseId='tu-kb-id',
retrievalConfiguration={
'vectorSearchConfiguration': {
'numberOfResults': 1 # Este es el parámetro correcto
}
}
You can see the documentation here