Turning Computer Vision and Real-Time Data into Day-of-Travel Decisioning
Executive Hook
1
I led the evolution of Lobby360 and Airport360 into an operational intelligence capability that fused computer vision, self-service telemetry, staffing data, and passenger-flow signals into a real-time decision layer for airport lobby and gate performance.
The Problem We Solved
2
Airport leaders often had no single, trusted view of what was happening across the lobby in real time. Kiosk usage, bag drop congestion, TSA queue conditions, staffing levels, group arrivals, and customer sentiment were all visible in fragments, but not in a form that could support immediate operational action.
That meant the operation was often managed reactively. Staffing adjustments came late, queue interventions were inconsistent, and frontline teams lacked a common picture of where friction was building. The challenge was not simply to visualize the lobby. It was to build a system that could convert live operational signals into decisions that improved throughput, customer experience, and staffing effectiveness on the same day.
Technology Deep Dive & Architecture
3
Lobby360 and Airport360 were architected as edge-plus-cloud operational intelligence platforms. On the edge, the video pipeline relied on HPE/NVIDIA-based compute running DeepStream inference pipelines to process real-time camera feeds and convert them into structured measures such as people count, queue conditions, kiosk dwell time, service rates, bag-drop performance, and customer-service wait times. Milestone video-management layer provided the camera stream interface, while camera-specific configuration and homography logic made it possible to translate raw video into operational metrics tied to each lane, kiosk, or service zone.
In the cloud, AWS was the natural landing zone for ingest, aggregation, and delivery. The architecture included capture outcome ingestion pipelines, APIs for kiosk telemetry, TSA lane status inputs, staffing data, and passenger-arrival signals, all feeding a backend that exposed dashboards, alerting, messaging and reusable APIs. The production stack combined Kinesis, containerized transform services, and a time-series plus relational analytics layer to expose near-real-time operational views. Use of horizon ML models to predict customer personas and likely in-person agent demand, using those predictions to support dynamic micro-assignments and staffing flex. This was not an LLM-first use case; it was a CV, telemetry, and predictive analytics system designed to become progressively more assistive as models matured.
AI Transformation Pattern
4
Lobby360/Airport360 sits in the middle of the enterprise AI transformation stack. Intelligent Airport created the sensor and data foundation. Lobby360 turned that foundation into operator-facing intelligence. From there, the capability could evolve toward forecasting, micro-assignments, and AI-assisted orchestration.
That progression matters. Enterprises do not jump from disconnected raw data to trusted AI actioning overnight. They build confidence by first making the signals visible, then operational, then predictive, and finally assistive. Lobby360 embodied that maturity curve by turning computer vision and telemetry into a real-time system of insight that frontline leaders could use.
Operational Resilience
5
The platform was engineered for live operations, not lab conditions. Edge processing reduced latency and dependency on constant round-trip cloud processing. Standardized infrastructure patterns, repeatable camera configuration, secondary-region planning, testability, and DR design all reflected production-grade thinking.
Just as important, the system was structured to support scale without bespoke reinvention. Each new station could follow a known implementation pattern across hardware, networking, inference configuration, data validation, and dashboard enablement. That combination of reliability and repeatability is what turns a pilot into an enterprise capability.
6
Cross-Functional Leadership
This initiative required close partnership across airport operations, digital product, advanced analytics, mobile teams, infrastructure, and external stakeholders. The work bridged frontline operational metrics, customer engagement, AI models, and the systems needed to distribute insight back into both employee and customer channels.
That cross-functional model is exactly what Enterprise AI transformation leadership requires. Success depended on aligning the people who owned the workflow, the teams who owned the data, and the technologists who could build the platform into a coherent operating capability.
Phased Rollout & Delivery
7
The delivery model was intentionally staged: prove the dashboard and use cases, harden one station for production, then build the scaling blueprint for multi-station rollout. This included not only application development, but also infrastructure dependencies, compute requirements, data-source validation, UX refinement, DevOps hardening, and operational readiness.
This reflects disciplined scale mechanics. It shows how to move from promising AI visualization to production-grade operational capability with a clear implementation pattern, funding logic, and rollout path.
What’s Next / The Legacy
8
Lobby360/Airport360 demonstrated how computer vision and operational telemetry can become a trusted, real-time decision layer for airport operations. It laid the groundwork for predictive staffing, dynamic micro-assignments, customer messaging, and broader conversational AI-assisted orchestration across the day-of-travel journey where leaders stay heads-up and signals are delivered via NLU.
This project is the bridge from platform foundation to operator adoption. It proves the ability to make AI useful where it matters most: in live workflows where better visibility and faster action improve both customer experience and operating performance.