We write the software that isn’t allowed to fail.

MTR stations run real-time crowd movement monitoring — escalator directions adjusted, congestion alerts raised — and a waiting-time indicator that predicts boarding waits from vision-based algorithms and machine learning. On the Airport Express, a single high-speed camera per platform estimates train seat availability across all 8 platforms and 35 km in under 25 seconds.

Built on EdgeAI Studio as a shared platform for every bureau and department — perception-based AI powered by cutting-edge machine learning algorithms, processed locally on specialised hardware. Their own teams pick ready-made models or train their own on CCTV, scanned documents and drone imagery. The Transport Department’s vehicle classification model reads 16 video streams, classifying vehicles into 15+ statutory classes for traffic surveys.

Shoppers photograph receipts in the MTR Mobile App, ELEMENTS app and UAS parking redemption — neural-network text detection reads mall name, payment type, date and total, a custom-built anti-fake module screens out forgeries, and each receipt completes end-to-end in about 7 seconds. Runs on EdgeAI Studio with custom vocabulary tuned to mall-specific text.

Gas risers corrode constantly inside tall buildings — and inspecting them traditionally means scaffolding the whole facade. Drones and stationary cameras film the risers, and an EdgeAI cluster runs three machine-learning models that detect pipes and corrosion areas, classify severity across 3 levels, and verify every finding — trained on thousands of hours of riser footage. Inspection reports carry verified defects, floor locations and a Riser Health Index instead of manual climbs.

Scaffolding safety checks traditionally mean climbing every structure by hand. TIE.Ai sensors strapped to the scaffolding stream wire-connection, temperature, tilt-angle and tension readings over MQTT into a React web portal — dashboard, scaffolding plan, per-sensor history, alerts and analytics — with the status grid refreshing every 5 seconds. Threshold breaches raise WhatsApp alarms to the site team straight away, so overloads and wire cuts are caught while the structure can still be secured.