For new Pega Knowledge Buddy Implementations they are now typically configuring the environment to use the AWS Titan embeddings model instead of the ADA 2 model. This will require you to change the similarity score of your Knowledge Buddy, which is found in the Advanced Settings tab. The recommended threshold value is 37. This adjustment aligns the retrieval behavior with the Titan model’s scoring characteristics and helps prevent empty or incomplete responses. For existing environments that are already configured with the Titan model, if Knowledge Buddy does not return responses, update the similarity score threshold to 37 and retest the configuration.
Thanks for calling this out — this is an important operational detail for teams using Knowledge Buddy.
Different embedding models have different similarity score distributions, so adjusting the threshold when moving to Titan is expected behavior, not a workaround.
What’s worth reinforcing is that embedding model changes should be treated as architecture changes, not just configuration tweaks. Retrieval quality, coverage, and user trust should always be revalidated after such changes.
If others have migrated between embedding models, sharing how they validated answer quality beyond threshold tuning would be very helpful.