I’m considering building a ChatBot on Google Cloud.
Initially, I wanted to use (and potentially train) an LLM on Google Cloud VertexAI.
The ChatBot should be trained with documents/web pages, etc… to build some knowledge.
I’ve briefly looked into the features that DialogFlow offers.
Dialogflow seems interesting in terms of knowledge base and entity recognition.
Would it make sense to use Dialogflow to understand users’ questions and get an answer, and then use an LLM to rephrase the answer in better English and potentially follow up the conversation?
Or do you think Dialogflow doesn’t offer anything over custom-trained LLM models?
I was thinking Dialogflow could be nice to restrict the parameter of the answers the ChatBot gives.
It is probably better to build a knowledge base with Dialogflow.
Dialogflow could prevent LLM from saying things it’s not supposed to.
What do you guys think?
Sequence diagram of the system