Construction Safety Incident Triage with OpenAI, Slack & Gmail

Instantly triage construction safety incidents—using OpenAI, Slack, and Gmail

When a safety issue is reported from your construction site, this n8n workflow automatically classifies the incident’s severity and OSHA recordability with OpenAI, checks subcontractor compliance via an HTTP API, and then routes alerts and corrective actions through Slack and Gmail—while logging every incident to Google Sheets.

What this workflow does

• Receives incident reports via webhook: Accepts a POST request to /safety-incident containing site ID, subcontractor, incident type, reported severity, and incident description.
• Normalizes the payload: Converts the incoming data into a consistent incident schema for reliable downstream processing.
• Uses OpenAI for triage: Classifies severity, determines whether the incident is OSHA recordable, and generates required corrective action guidance.
• Looks up subcontractor compliance: Fetches compliance details from a compliance database using an HTTP API.
• Routes by AI-assessed severity:
  • Critical: Posts to a Slack safety leadership channel.
  • Critical: Creates a draft OSHA/regulatory report in the compliance system and emails the subcontractor corrective actions.
  • Moderate/Minor: Emails the subcontractor corrective actions and logs the incident to Google Sheets (including severity and recordability).

Use cases

• Safety coordinators submit incident reports from any internal tool or form that can send a webhook.
• General contractors need consistent OSHA recordability decisions and faster subcontractor corrective action communication.
• Automation engineers want an auditable incident register backed by OpenAI triage and Slack/Gmail notifications.

Technical details

• n8n nodes/integrations: webhook, set, switch, OpenAI (gpt-4o-mini), Slack, Gmail, sticky note, and HTTP API.
• Setup: Enable the webhook endpoint by copying the n8n Webhook URL and sending incident reports to /safety-incident via POST. Add an OpenAI credential and keep the model selection (gpt-4o-mini) or update it if needed.