- Python
- time series forecasting
- LLM IVR
- RAG chatbot
Cutting an air cargo forecast cycle from six weeks to minutes
A set of LLM and ML systems for freight forwarding and airport operations: multi-granularity tonnage forecasting that replaced a six week finance cycle, an LLM powered IVR that cut routine support calls by 30 to 40 percent, and a real time dock and airport management chatbot.
Problem
Air cargo operators ran three separate bottlenecks off the same data. The finance team spent six weeks producing a tonnage forecast. Routine customer inquiries went to human agents by phone, and freight forwarders and airport authorities made cargo management decisions without a live view of the dock.
Approach
Three systems, one industry, built over the same airport, cost centre, customer, and commodity level data. Multi-granularity time series forecasting for tonnage. An LLM powered IVR for routine inquiries. A real time airport and dock management chatbot for freight forwarders and airport authorities.
Architecture
Tonnage forecasting, the LLM powered IVR, and the operations chatbot sit side by side over a shared airport, cost centre, customer, and commodity level data layer.
The hard part
Tonnage had to forecast at four granularities at once: airport, cost centre, customer, and commodity. Finance does not want a single number, it wants the number it can defend at whichever level it is asked about. That is the difference between a model and something the finance team will actually use to replace a six week cycle.
Outcome
The forecast cycle went from six weeks to minutes. The IVR automated routine customer inquiries and reduced support call volume by 30 to 40 percent. The chatbot improved cargo management decisions for freight forwarders and airport authorities, and together the three read as a suite rather than a single build.
Next