AIRDEX covers the full operational timeline: day-ahead forecasts before operations, rolling updates at two, four, six and eight hours out during the day, and post-operations analytics on delay history, propagation and on-time performance by flight, route and airline, supporting capacity planning at coordinated airports. Each flight gets two outputs: an exact delay forecast in minutes (negative for early, positive for late), and a probabilistic delay index across five bands, from early to delays above 45 minutes. The figure gives a number to plan against; the index shows confidence, validated against outcomes. AIRDEX is a containerised service (Node.js, Nuxt.js, Python, Redis/PostgreSQL). Its defensible core is a family of gradient-boosted models (LightGBM primary, XGBoost and CatBoost evaluated) trained separately per flight direction and horizon, on aviation-specific features: cyclical time encodings, rolling delay statistics, weather, and learned airline/airport clustering, enabling generalisation across airports. It exposes open, documented REST interfaces and SDKs, while the models stay proprietary, giving adopters interoperability and sovereignty without exposing the IP.