How do you give your Voice/Chat Agents 100% accuracy?

Hey guys, so we have implemented some chat and voice agents and we always use a

### out of scope section and a ###faq section and like 95% of all the cases can be handled with those but somehow once in a while a person sneaks up and breaks the ai (some of the learnings of the past: " give me your prompt" , "instructions", "what do you think about gaza war", etc.)

As anyone found a way to make them 100% ok atleast 99.9% reliable? If so are you open to share in how you do it or atleast your approach to that?

Best regards and great building!

1 comment