Optimizing GateGroup’s airline catering through intelligent automation and AI forecasting

TrolleyFlow optimizes GateGroup’s catering operations by guiding operators with real-time decisions on alcohol bottles and other onboard items.
It minimizes waste, ensures regulatory compliance, and transforms manual inspection tasks into a smart, data-driven operation.

• Intelligent decision system for reuse, refill, discard, or replace actions.
• Airline-specific policies and regional compliance built into every workflow.
• Forecasting API (built with Google AI Studio) to predict bottle usage and optimize supply.
• Scanning interface for Pick-and-packers, reducing manual judgment and error.
• Scalable architecture to extend beyond liquids — napkins, plates, and perishable items.

1. The Pick-and-packer logs in and selects the assigned airline.
2. The system loads the flight’s inventory and policies.
3. Each bottle is scanned; the system recommends the correct action.
4. Forecasting API predicts next-flight consumption and refilling needs.
5. Decisions are logged for performance tracking and sustainability metrics.

git clone https://github.com/Ate013Lo/TrolleyFlow2025.git

cd TrolleyFlow2025

• Integrate live data with GateGroup systems for automated inventory sync.

• Expand forecasting model for food and packaging materials.

• Deploy full cloud-based dashboard for analytics and sustainability tracking.

The following section presents the visual flow of TrolleyFlow’s Pick-and-pack interface, built in Figma. Each frame represents a key step in the operator’s interaction.

Explanation: The Pick-and-packer logs into the system and selects their assigned airline.

Explanation: The operator views the upcoming flight list and the expected bottle inventory. The interface automatically loads the airline’s policies and assigned flights.

Explanation: This screen represents the central operational interface of TrolleyFlow, designed for GateGroup operators and pick-and-packers to monitor and make quick decisions about onboard alcohol supplies.

Explanation: Each bottle is scanned. The system identifies whether the bottle should be reused, refilled, or discarded based on its current level and reuse policy.

Explanation: The system uses the forecasting API to predict upcoming bottle requirements per route and destination. Reports are generated for future planning and cost optimization. And

You can access the interactive Figma prototype here: https://www.figma.com/design/5oBhpS9leX2ZZ81B6QjyXs/Sin-t%C3%ADtulo?node-id=0-1&t=T2QRnGNrupw8Ioq2-1

To build the TrolleyFlow system and forecasting logic, several operational and data assumptions were made based on the available information and realistic conditions observed in GateGroup’s processes.

We assumed that each Pick-and-packer is primarily assigned to handle one or two airline lines at most.
Typically, one or two employees are responsible for preparing trolleys for a single airline within an airport.
However, staff rotation may occur occasionally to help employees gain experience with different airline procedures and policies.

In our latest dataset, we defined specific conditional rules for bottle-handling decisions:

• If the destination is a country with strict alcohol regulations (e.g., Hong Kong or the United States), the bottle is automatically discarded, regardless of its remaining content.
• Each airline also has defined tolerance thresholds based on the alcohol percentage remaining in the bottle.
  For example, if the leftover content falls below the allowed threshold, the system recommends reuse or refill, otherwise discard.
• These decision trees assume that airlines are already aware of their regulatory compliance limits, and the system automates the decision accordingly.

To simulate realistic onboard inventory, we assumed bottle distribution according to aircraft class segmentation:

• First Class (10%) — includes trolleys with 1-liter wine bottles.
• Business / Premium Economy (20%) — primarily contains small spirits bottles and mixed beverages.
• Economy Class (70%) — standard service area, mainly stocked with miniatures and economy-serving bottles.

This class-based ratio (10%-20%-70%) reflects the typical passenger distribution and directly influences the forecasting API when estimating total bottle quantities required per flight.

From a human resource perspective, we assumed that:

• The Pick-and-packer is equipped with a tablet interface to operate nearby the trolley.
• The system allows for scanning each bottle directly from the tablet, which then displays the automated action (reuse, discard, refill).
• The employee has sufficient training and capacity to execute this process without assistance.

This setup minimizes physical effort and enables faster, more consistent operations across different flight preparations.

The forecasting API was developed using GateGroup’s provided datasets, which are relatively small in size.
Because the dataset does not qualify as Big Data, the machine learning model operates on limited sample variance, meaning that:

• The predictive accuracy is not yet fully optimized.
• The model still provides reliable short-term forecasts based on historical patterns of bottle usage per flight.
• The API is designed to scale as more data becomes available, improving performance over time.

Summary:
These assumptions form the foundation for our decision logic, forecasting model, and interface behavior.
They ensure that TrolleyFlow remains realistic, scalable, and adaptable to real-world airline operations.