I am trying to make a chat bot for my enterprise. Creating FAQ bots is easy, using RAG pipeline.
But coming to decision making, for example i.e. processing refunds, checking order status, payment status, etc. How to do that? Shall I just create agent for each task and classify intent of user and call them? or is there any other method?
1 Like
Have you tried OpenAI actions with Zapier like here https://youtu.be/blTSl1z9XYU . This is a demo of connecting Shopify api with a chatbot. You can do the same with any other api as well
It’s basic wrapper for RAG, I need more of a decision making based on user data from api
Why not use the Assistant API with function calling ? In each function you not only describe what it should do but the conditions for it to be ran, so that the model can decides when to process a refund for example.
yes, I am thinking on same line. Is there any workout example for multiple cases?
Just use normal function calling and define a separate function per task. Parallel function calling is now possible with 1106 models so there is no need imo to add a separate function to decide which function to call. You can just add all those functions and let the AI handle everything.