Briefing review support
ModeledFaster scanning
The system was designed to reduce review time by converting dense NOTAM text into structured, pilot-friendly summaries.
AI Operational Summarization
SkyGuard — AI NOTAM Summarization for Pilots
An AI briefing workflow that converts complex NOTAMs into structured, pilot-friendly summaries with conservative risk classification and human review.
Sanitized case studyMay 30, 2026
SkyGuard is an AI briefing workflow that converts complex NOTAMs into structured, pilot-friendly operational summaries with conservative risk classification and human verification.
LLMStructured SummarizationRisk LabelsReview WorkflowPrompt Constraints
Problem
NOTAM review can be time-consuming and cognitively heavy. The danger is not only volume; it is missing a critical operational detail or misunderstanding relevance under time pressure.
Users
The intended users were pilots and operational aviation users who needed faster briefing review while still relying on official NOTAM sources.
My role
Constraints
- NOTAMs are safety-sensitive and must be verified against official sources.
- Summaries can be misleading if they omit a qualifier, location, time window, or operational condition.
- The system needed conservative wording and explicit boundaries.
- Human review was mandatory.
System built
The workflow ingested NOTAM text, transformed it into structured summaries, applied risk labels, grouped information into pilot-friendly sections, and preserved the need for verification.
- 01NOTAM intake
- 02Normalization
- 03Structured summarization
- 04Risk labeling
- 05Route or context grouping
- 06Pilot-friendly brief
- 07Human verification
Architecture snapshot
SkyGuard treated summarization as a constrained transformation, not a free-form answer. The output emphasized scope, operational relevance, and warnings where uncertainty remained.
AI/ML approach
The AI layer used structured prompting, output constraints, risk categories, and conservative language. It was designed to reduce review effort without becoming a source of truth.
Product decisions
- Use careful wording and avoid overstating certainty.
- Separate summary from verification requirement.
- Highlight high-risk or time-sensitive items.
- Keep output structured for scanning.
- Avoid hidden decisions that could imply automation.
Evaluation & reliability
01 · Grounding
Summaries needed to map back to original NOTAM text and preserve operational qualifiers.
02 · Failure modes
Key risks included missed critical information, incorrect risk classification, compressed nuance, and stale source data.
03 · Human review
Every output required verification against official NOTAMs.
04 · Operational risk
A wrong summary could create safety risk, so the product boundary had to be explicit.
05 · Adoption
Usefulness depended on reducing cognitive load while maintaining trust.
06 · Production readiness
A production version would need structured eval sets, logging, versioned prompts, and review workflows.
Impact
Limitations
What I learned
Safety-sensitive summarization requires product restraint. The system must help users notice and verify, not encourage blind trust.
What I would improve next
- Build a structured evaluation set of NOTAM examples.
- Add route-specific filtering and relevance checks.
- Collect pilot feedback on wording and categorization.
- Track missed critical information and classification errors.
Expandable technical notes
The most important technical choice was constraining the output format. A production version would preserve source references, version prompt templates, log summary decisions, and compare generated briefs against a curated evaluation set with known critical items.