How to build a Safety Ops Chatbot

Workflow Overview

1. User Input Collection

• User Question (Input Node)

  • The user submits a safety-related question (e.g., “What steps should I take to safely clean up a small chemical spill in the workplace?”).

  • This is the main entry point for the workflow.

• File Upload (Files Node)

  • The user can upload supporting files, such as screenshots or logs, to provide additional context for their question.

  • These files are processed (including OCR for images) so their content can be referenced later.

2. Reference Material Gathering

• OSHA Injuries and Incidents (Web Page Node)

  • The workflow pulls in content from the OSHA Severe Injury web page as a reference for incident data and best practices.

• OSHA Safety Standards (Knowledge Base Node)

  • Searches a curated set of OSHA policy documents (e.g., PDFs) for relevant standards and procedures.

  • Only specific files are indexed for search, ensuring focused, authoritative results.

• Safety Risk Register (Knowledge Base Node)

  • Searches your organization’s OneDrive for internal risk registers or related safety documentation.

• Safety, Health & Environment Management (Knowledge Base Node)

  • Searches your organization’s Google Drive for internal safety management system manuals, policies, and procedures.

3. AI-Powered Answer Generation

• Anthropic Agent 1 (LLM Node)

  • This is a large language model (Claude 4.5 Sonnet) acting as a virtual Safety Manager mentor.

  • It receives:

    • The user’s question,

    • Any uploaded files,

    • Content from the OSHA web page,

    • Results from all three knowledge base searches (OSHA standards, internal risk register, internal safety management system).

  • The AI is instructed to:

    • Prioritize company policies first, then supplement with OSHA and other standards.

    • Provide actionable, practical, and compliant advice.

    • Cite relevant standards or policy sections in its response.

    • Escalate to the safety team if the question cannot be answered.

4. Output to User

• User Output (Output Node)

  • The AI-generated answer, complete with references and citations, is displayed to the user.

How the Data Flows

1. User submits a question (and optionally uploads files).

2. The question and files are sent to:

   • Internal and external knowledge bases (for searching relevant documents).

   • The OSHA web page (for additional context).

3. All gathered information is fed into the AI agent, which synthesizes a comprehensive, standards-compliant answer.

4. The answer is shown to the user in a clear, actionable format.

Visual Node Map

Summary

This workflow is designed to provide users with authoritative, actionable safety guidance by combining:

• User input,

• Uploaded evidence,

• External regulatory standards (OSHA),

• Internal company policies and risk registers,

• And expert-level AI reasoning.

The result is a robust, compliant, and context-aware answer to workplace safety questions.