AI customer support using RAG
(RAG AI AGENT) Responding to repeated tickets sure takes a lot of time and human labour which can be used way more productivly elsewhere.
Today's workflow will automate this very same task but not like the one I made previously but, by using RAG instead.
About the workflow:
-> This one is a set of two workflows, one for storing the repeated tickets in a vector database and the other for answer the query of clients and sending the unique/new ones over to the team via Slack.
Data Ingestion Workflow :
-> When triggered manually(can be triggered on schedule or even when the data in google drive is updated automatically), the drive file storing data is extracted into the workflow.
-> This drive is then stored into the pinecone vector database using an embedding model and a text splitter as a Data Loader.
Query Responder:
-> Triggered when email received.
-> Goes to the Ticket Responder RAG Agent which is connected to an LLM model and the vector database storing the answer for repeated queries asked by most clients.
-> The output is in text so output parser node used to turn it into JSON schemas.
-> Based on the query ticket, the if node checks if needed to be escalated to the team or to be replied without escalating.
-> Escalated tickets are sent over to a Slack Channel where the query team can resolve the new/unique tickets and even update the database about new queries.
Time to build : 50 min
Please drop your suggestions about this workflow and feel free to reach out for JSON file.
I thought building a RAG agent will be tough but this one was the most easiest project I've ever built on n8n.
Today's workflow will automate this very same task but not like the one I made previously but, by using RAG instead.
About the workflow:
-> This one is a set of two workflows, one for storing the repeated tickets in a vector database and the other for answer the query of clients and sending the unique/new ones over to the team via Slack.
Data Ingestion Workflow :
-> When triggered manually(can be triggered on schedule or even when the data in google drive is updated automatically), the drive file storing data is extracted into the workflow.
-> This drive is then stored into the pinecone vector database using an embedding model and a text splitter as a Data Loader.
Query Responder:
-> Triggered when email received.
-> Goes to the Ticket Responder RAG Agent which is connected to an LLM model and the vector database storing the answer for repeated queries asked by most clients.
-> The output is in text so output parser node used to turn it into JSON schemas.
-> Based on the query ticket, the if node checks if needed to be escalated to the team or to be replied without escalating.
-> Escalated tickets are sent over to a Slack Channel where the query team can resolve the new/unique tickets and even update the database about new queries.
Time to build : 50 min
Please drop your suggestions about this workflow and feel free to reach out for JSON file.
I thought building a RAG agent will be tough but this one was the most easiest project I've ever built on n8n.
