How to build a Customer Support Chatbot

Workflow Overview

1. Customer Question (Input Node)

• Purpose:
  This is where the user (customer or support rep) enters their question or issue.

• Node: in-0 (Customer Question)

2. Knowledge Search (Parallel Retrieval)

• Purpose:
  The workflow searches for relevant information to answer the question using two sources:

  • A. Search Knowledge Bases (Action Node):
    Uses the "Search Knowledge Bases" action to query a specific knowledge base (configured in the node) with the customer’s question.

  • B. SharePoint 2 (Knowledge Base Node):
    Searches a SharePoint knowledge base for relevant content.

• Nodes:

  • action-0 (Search Knowledge Bases)

  • knowledgebase-2 (SharePoint 2)

• How it works:
  Both nodes receive the customer’s question as input and independently search their respective sources.

3. AI-Powered Answer Generation

• Purpose:
  An AI model (OpenAI LLM) combines the original question and the retrieved knowledge base results to generate a helpful, concise answer.

• Node: llm-1 (OpenAI)

• How it works:

  • The LLM receives:

    • The customer’s question

    • Results from both knowledge search nodes

  • It generates a brief answer, asks for clarification if needed, and proposes follow-up questions.

  • The LLM also provides citations for the sources it used.

4. Answer Output

• Purpose:
  The generated answer is displayed to the user.

• Node: out-0 (Answer)

• How it works:
  The output from the LLM is sent directly to this node for user viewing.

5. Airtable Instruction Generation

• Purpose:
  The workflow prepares a natural language instruction for Airtable, summarizing the question and answer for record-keeping or follow-up.

• Node: llm-2 (OpenAI 2)

• How it works:

  • Receives:

    • The original question

    • The answer generated by the first LLM

  • Converts these into a clear instruction for Airtable, following a specific format.

6. Write to Airtable

• Purpose:
  The workflow writes the instruction (from the previous step) into a specified Airtable base and table.

• Node: action-1 (Write to Airtable)

• How it works:

  • Takes the formatted instruction from the second LLM

  • Uses the Airtable connection and table details to create a new record

Node-by-Node Summary

How the Flow Helps

• Automates customer support:
  Answers questions using both internal and SharePoint knowledge.

• Records interactions:
  Summarizes and logs each Q&A in Airtable for tracking, analytics, or follow-up.

• Ensures traceability:
  Citations are included for transparency and compliance.