Significant Response Lag with Streaming Avatar Using Custom Knowledge Base

Hi everyone,
I'm currently exploring Heygen Labs' Streaming Avatar and integrating it with a custom knowledge base I created. While the setup and integration through the SDK went smoothly, I've been facing a noticeable lag in the avatar's responses during testing.
I'm trying to understand if this delay is expected under certain conditions, or if there are recommended optimizations I can apply—either within the SDK, my knowledge base structure, or the hosting environment—to reduce this latency.
Has anyone else experienced similar issues? Are there best practices or configuration tweaks that can help improve response time?
Any insights or guidance would be greatly appreciated. Thanks in advance!
Best,
Sai