Quick Overview
This workflow receives vessel telemetry via a webhook, runs rule-based data quality and anomaly scoring, optionally uses Anthropic Claude and OpenAI for diagnosis, forecasting, and reconstruction, then routes clean data to downstream HTTP endpoints while logging and alerting issues via Google Sheets and Gmail.
How it works
- Receives a vessel telemetry reading via a POST webhook request.
- Validates and cleans sensor fields, detects missing/out-of-range/stale/flatline/GPS-jump inconsistencies, and calculates data quality, anomaly, risk, and severity scores.
- Responds immediately to the webhook caller with a compact quality verdict (status and scores).
- Builds a short rolling history per vessel and, when risk exceeds the configured threshold, uses Anthropic Claude agents to diagnose the anomaly, forecast near-term failure risk, reconstruct plausible substitute values (via OpenAI), and generate a crew safety advisory.
- Composes a final event record (including diagnosis metadata) and routes it based on data status: trusted and degraded events are posted to the configured downstream HTTP endpoint, while rejected events are quarantined in Google Sheets.
- Sends a Gmail alert for high/critical events and appends any non-trusted or non-none-severity events to a Google Sheets Events log for later analysis.
Setup
- Configure the webhook source system to POST telemetry to the workflow’s webhook URL (path: /vessel-data-ingestion).
- Add credentials for Anthropic (Claude), OpenAI, Google Sheets OAuth2, and Gmail OAuth2.
- Create/select a Google Sheets document with tabs named "Events" and "Quarantine", and set the target document/sheet settings in the two Google Sheets nodes.
- Ensure each incoming payload includes a vesselId plus the expected sensor fields (for example speedKnots, rpm, fuelRateLph, engineTempC, latitude, longitude) and include config values such as limits/thresholds, alertEmail, and downstreamAgentsUrl (the HTTP endpoint that receives downstreamPayload).