How to build an IT Chatbot

Workflow Overview

1. User Question (Input Node)

• Purpose:
  Collects the user's IT support question (e.g., “My VPN won’t connect”).

• How it works:
  The user types their question, which is then passed to the rest of the workflow.

2. File Upload (Files Node)

• Purpose:
  Lets users upload screenshots or log files to help diagnose IT issues.

• How it works:
  Users can attach files (images, logs, etc.), which are processed (including OCR for images) and made available for troubleshooting.

3. IT Policies (Knowledge Base Node)

• Purpose:
  Searches your organization’s Google Drive for IT-related files, guides, or troubleshooting documents.

• How it works:
  The user’s question is used to search selected folders (like “IT Policies”) in Google Drive, retrieving relevant documents or policies.

4. OpenAI (LLM Node)

• Purpose:
  Generates an IT support answer by combining all available information.

• How it works:

  • Receives the user’s question, uploaded files, and search results from the knowledge base.

  • Uses a system prompt instructing it to act as an IT support agent, referencing both uploaded files and knowledge base content.

  • If the answer cannot be found, it is set up to send an email to the appropriate contact (though the email action is referenced in the prompt, not as a separate node).

5. User Output (Output Node)

• Purpose:
  Displays the formatted IT support answer to the user.

• How it works:
  The answer generated by the LLM is shown to the user, including any relevant citations from the knowledge base or uploaded files.

How the Data Flows

1. User submits a question →
   Sent to both the File Upload (for context) and IT Policies (for document search).

2. User uploads files (optional) →
   Files are processed and made available for the LLM.

3. IT Policies node searches Google Drive →
   Returns relevant IT documents or policies.

4. LLM node receives:

   • The user’s question,

   • Uploaded files,

   • Knowledge base search results,
     and generates a support answer.

5. Output node displays the answer to the user.

Key Features

• Multi-source context: Combines user input, uploaded evidence, and internal documentation.

• Automated support: Uses AI to generate answers, referencing policies and user-provided files.

• Fallback: If the AI cannot answer, it is instructed to send an email for human follow-up (though this is only referenced in the LLM prompt, not as a separate automated action node).