During a site visit to Delta Airlines, I noticed something unexpected: a wall-mounted screen cycling through mugshots. Security guards were manually scanning faces against a printed "banned visitor" list.
They were frustrated, inefficient, and overwhelmed. It was like expecting Google-level results from a flipbook. That is when I had an idea.
The Suggestion: "Describe the Person Instead"
I proposed an AI-based image matcher that could take verbal traits (male, mid-30s, beard, dark hoodie) and return a ranked list of likely matches. Not exact, but close enough to help a human decide faster.
I worked with our product team to create a basic prototype using:
- 👁️Face encoding via AWS Rekognition
- 🏷️Tag-based search using attributes from prior incidents
- 📊Confidence thresholds based on field validation
"This feels like Iron Man's HUD for security. Can we roll this out to all entrances?" — Delta Facilities Director
Business Impact
What started as a one-off suggestion turned into a formal PoC. Delta piloted it at two gates with measurable results: a 60% reduction in guard check time and higher incident capture rates.
The client felt heard, and the field teams were finally empowered, not just monitored.
SEs do not just answer questions. They spot the unasked ones.
SE Takeaways
- 🏢Field visits matter. Some pain points never show up in Jira.
- 💻Prototypes beat explanations. Even a basic UI gets buy-in faster than a spec sheet.
- 🔍SEs do not just answer questions, they spot the unasked ones.
More from the SE Blog