User Story: Prevent Hallucinations

[givecoffee]

As an internal user, I want the chatbot to refuse answering questions that are outside the knowledge base so that false information is not received.

Criteria to meet this:

• Implement confidence threshold and ONLY answer if RAG retrieval score exceeds 0.5-0.7 (or adjust as needed)
• Return an explicit message "I do not have information about that, please contact xyz directly"
• Log all refused queries to identify knowledge gaps
• Add test cases for questions prone to hallucination (general knowledge, subjective or opinion based, etc)