Shift Handover Report Synthesis and Anomaly Highlighting Automation

Engineers currently spend the first hour of a shift manually compiling notes from logs, tickets, and telemetry dashboards, and the last hour writing a summary for the next team. This automation ingests structured and unstructured data from systems like Satellite Toolkit (STK) telemetry databases, Jira/ServiceNow tickets, and Slack/Microsoft Teams channels, using an LLM orchestrator (e.g., LangGraph) to synthesize a concise, natural-language summary. The result is a consistent, auditable handover document generated in under 60 seconds, reclaiming engineering time for anomaly investigation and proactive planning.

Manual handovers are prone to omission bias, where non-critical but contextually important events are left out. This workflow applies a multi-agent review layer: one agent extracts key events, another scores anomaly severity against historical baselines, and a third highlights open action items from the ticketing system. By enforcing a consistent review framework across all data sources, the automation ensures no critical open ticket or lingering subsystem drift is missed, measurably improving the incoming team's readiness and reducing the risk of oversight.

When an anomaly is detected mid-shift, it's often buried in logs. This workflow includes an anomaly highlighting and prioritization engine that correlates real-time telemetry alerts with the shift's event log. High-severity items are prominently featured in the summary with recommended actions and links to relevant SOPs or diagnostic dashboards. This direct routing cuts the time for the new team to contextualize and act on ongoing issues by over 50%, directly impacting mission continuity and subsystem recovery times.

Beyond daily operations, automated handovers serve as a searchable, timestamped record of fleet status and crew decisions. The workflow archives each report with metadata tagging (satellite IDs, anomaly codes, responsible engineers) in a system like Elasticsearch or a dedicated data lake. This creates a defensible audit trail for post-incident reviews, regulatory inquiries (e.g., FCC filings on satellite health), and training new operators, reducing compliance overhead and litigation risk.

As a constellation grows from dozens to hundreds of satellites, manual handover processes break down. This custom architecture is designed for scale: the orchestration layer (e.g., Apache Airflow or Temporal) manages the ingestion pipeline, the LLM agents are containerized for elastic scaling, and the output format is consistent regardless of fleet size. This allows a single operations team to manage a significantly larger asset base without degrading handover quality or requiring proportional staff increases, directly improving operational leverage.

Implementation risk is minimized by designing the workflow as an overlay to existing systems, not a replacement. The build uses APIs to pull from COSMOS/MOOS-IvP for telemetry, ServiceNow for tickets, and Confluence for SOPs. A pilot can be deployed in a staging environment within 3-4 weeks, using historical data to refine the summarization and highlighting logic. This phased approach delivers immediate value, proves ROI, and builds confidence for a full production rollout across all control centers.

Automating the 30-60 minute manual report-writing task for each 8-12 hour shift directly reduces operational labor costs. More critically, it ensures the incoming team receives a comprehensive, unbiased summary of all subsystem statuses, resolved issues, and open anomalies, eliminating the risk of critical context being lost in a verbal handover. This improves decision velocity at shift start and supports scaling operations without proportionally increasing headcount.

The workflow is built on a LangGraph or similar orchestration layer that ingests structured and unstructured data from multiple sources in parallel: telemetry databases (e.g., InfluxDB), event logs (Splunk), ticketing systems (Jira, ServiceNow), and operator notes. A primary LLM agent synthesizes this data against a predefined template, using RAG over historical reports and mission procedures to ensure contextual accuracy. A secondary agent cross-references telemetry against statistical baselines to highlight anomalies mentioned in the summary.

Implementation begins with a pilot on a single satellite or orbital plane. Phase 1 automates data aggregation and draft generation, requiring operator review and sign-off before the report is finalized. Phase 2 introduces automated anomaly highlighting based on configurable thresholds. The final phase integrates the approved report directly into the daily operations plan and mission control dashboard, with a seamless override mechanism for engineers to append notes or corrections.

The system's reliability depends on upstream data quality. Validation agents check for missing telemetry streams or stalled ticketing queues before synthesis begins, triggering alerts to the outgoing shift if gaps are detected. The final report is versioned and stored with a complete audit trail of source data. Approval is managed through a simple web interface or integrated into the existing mission control software, requiring at least one outgoing shift engineer to validate and release the summary.

The workflow does not operate in isolation. It must integrate with:

• Telemetry & Event Systems: To pull time-series data and log entries.
• Ticketing & Task Management: To list open and recently closed actions.
• Orbital & Planning Databases: To provide context on upcoming events (eclipses, maneuvers, ground contacts).
• Mission Control Dashboards: To publish the final report as the primary handover artifact.
• Alerting Systems: To flag if the report generation itself fails.

During major anomalies or contingency operations, the automated report serves as a baseline. Protocols define when operators must switch to a dedicated contingency reporting mode. The system includes exception routing for low-confidence summaries—such as when data conflicts are detected—escalating these directly to a shift lead for manual resolution. Regular reviews of report quality and highlighted anomalies ensure the LLM agents and anomaly models remain aligned with evolving operational priorities.