I'm kinda curious how you back yourself from hallucinations or uncontrolled behaviour of your chatbots ? Do you use some kind of SLA ?
Anyway good work !
It's important to reduce hallucinations, and with a good prompt it's possible to reduce them.
You basically tell the LLM to avoid answering if the context is not enough.
So, most of the time the Chatbot will correctly reply, and that's already a big advantage.
With Userdesk you can supervise the conversations, and the Chat widget automatically asks for the email of the user, so you can always provide support via the live chat or email later on.
It's important to reduce hallucinations, and with a good prompt it's possible to reduce them.
You basically tell the LLM to avoid answering if the context is not enough.
So, most of the time the Chatbot will correctly reply, and that's already a big advantage.
With Userdesk you can supervise the conversations, and the Chat widget automatically asks for the email of the user, so you can always provide support via the live chat or email later on.