Hello everyone! I am the admin for our German organization and we are very excited to be exploring the new Agent Assist features! We want to implement a semantic search engine that can find relevant knowledge articles for our agents even when the customer uses different terminology than what is in our documentation. I am looking at using Amazon OpenSearch and Bedrock for the vector embeddings. Has anyone successfully integrated an external semantic search engine with the Genesys Cloud Agent Assist framework?
Greetings Nao82. I am a migration consultant and I have seen many clients ask for this. You can integrate an external search engine by using the ‘Knowledge Base’ integration in the AppFoundry. However, for a custom Bedrock and OpenSearch setup, you will likely need to build a middleware service. Your middleware receives the interaction events via the Notification API, generates the embeddings using Bedrock, searches OpenSearch, and then pushes the results back to the agent’s workspace using a custom interaction widget.
I am a healthcare developer and we have very strict security requirements for our search results. Nao82, if you use Bedrock, make sure you are not sending any PII in your search queries to the AI model. We found that we had to implement a local redaction layer in our middleware before we sent the customer utterances to the embedding service. Also, the latency for generating embeddings in real time can be an issue. You might want to cache common search results to keep the agent experience fast.