How Proactive Q&A Works During Run Time
The Proactive Q&A leverages an LLM service to generate questions in real-time.
Service Initiation - When a new session begins, Real-Time Guidance Agent triggers the Proactive Knowledge Service with the conversation ID, Template ID, and Version. The service then retrieves new turn transcripts from the conversation service.
Question Filtering - To avoid repeated suggestions, the Question Necessity Classifier LLM component ensures that subsequent turns in the same session do not generate duplicate questions. It accomplishes this by saving and comparing previously generated questions.
LLM Question Generator - A GPT/LLaMA-based model converts each conversation turn into a clear, standardized question.
Semantic Similarity Model - A composite similarity algorithm compares the generated question with the Q&A cache associated with the Template.
Cache Bank - A collection of Q&A pairs associated with a template. The Proactive Knowledge Service checks the cache first before calling KaaS.
Threshold Check - The generated question is compared with the questions stored in the Q&A cache or knowledge base. Based on the predefined similarity threshold, a matching result is determined.
Above Threshold - Return the best cached answer.
Below Threshold - Use KaaS for a more accurate answer.
Cache Update - If KaaS is called, the new Q&A pair is added to the cache for future use.
Final Output - A single Q&A is displayed non-intrusively in the AI Agent Assist application.
