How Streaming ML saved Christmas: a Helicon use case for Santa’s delivery optimization

This year we have had the lifetime opportunity to team up with Santa’s team in Rovaniemi. Santa’s Office CEE (Chief Executive Elf) asked us to devise a solution for maximising the efficiency of sleigh-based delivery, in order to guarantee a safe and effortless distribution of gifts all around the world.

We immediately thought of using technology to monitor in real time sleigh and reindeers’ performance, leveraging data to dynamically adjust the gift-delivery strategy. This can be achieved thanks to the combination of connected telemetry sensors and machine learning models to generate descriptive and prescriptive intelligence.

Our Solution

Data Collection

Since the year 2000, Santa has started collecting a vast range of gift-delivery data thanks to a sensor-based telemetry system on his sleigh and reindeers.

Data collected can be grouped into 3 categories:

• Environment:
  • Wind Speed
  • Temperature
  • Visibility
  • Precipitation
  • Altitude above sea level
  • Oxygen in the atmosphere

• Delivery place conditions:
  • Presence of chimney
  • Type of house
  • Presence of in-house pets
  • Distance from next delivery place

• Sleigh:
  • Balancing of transported cargo
  • Weight of transported cargo
  • Reindeer energy
  • Wear of runners
  • Speed
  • Temperature on board
  • GPS

The two main challenges to overcome for Father Christmas are:

1) Assess when he will be able to safely reach the following point of delivery;
2) Estimate when he will be able to complete all deliveries.

Thanks to the historical data collected in 22 years, from Christmas 2000 until 2021, Santa Claus was able to label the two targets. This allowed us to create a supervisioned multi-target AI model that is capable of accurately predicting in real time the necessary delivery time. In this way, Santa can dynamically adjust his delivery speed based on the model’s output.

Our study on Santa’s Sleigh and reindeers

Data Exploration

Following Exploratory Data Analysis, it can be seen that all variables collected have an influence on the target. For instance, if Santa’s sleigh is flying at high altitude, the reindeers can experience shortness of breath and thus reduce the speed. If they are tired and flying on low energy levels, they need to stop to rest and eat. If the delivery point lacks a chimney, or a watchdog is in place, dropping the cargo may be more difficult and thus time consuming.

Deep Learning Model

After exploring the variables, we worked with Santa to develop a deep learning mode, called Christmas-Net-V1, that was trained on the huge amount of available historical data. The model delivered outstanding performance with the two targets:

• Prediction of next gift delivery time with 1,4 μs MAE (Mean Absolute Error)
• Prediction of time remaining for delivery completion with 15 μs MAE (Mean Absolute Error)

Helicon to enrich real-time data with AI (and save Christmas)

We used Helicon to combine the streaming data from Santa’s sensors and the Christmas-Net-V1 model. Helicon is a codeless Data for AI & MLOps solution that allows data teams to deploy, serve and monitor ML models effortlessly, driving productivity and reducing time to market.

The platform core lies in the streaming, MLOps and pipelining capabilities. The first concerns the collection of real-time data from external sources such as IoT & telemetry systems; the MLOps section enables the model serving and monitoring; the visual pipelines editor defines the operations, both preset and custom, that are to be performed on the flowing data.

In our Christmas use case, data collected by sensors are fed into a Kafka topic as soon as they are generated, to be then processed by a pipeline that moves data forward and transforms them in the way required by the model input.

Finally, a series of webhooks alert in real-time the on-board navigation system when certain conditions are met and Santa must revise the delivery route by accelerating, making a detour, etc.

Conclusion & Next Steps

We are proud to have worked with Santa and his team to develop a reliable solution to ensure a timely gift delivery and increase the Christmas spirit. This year, event stream processing and machine learning will play a role in making children’s wishes come true.

We will keep on evolving our Helicon-powered solution to include more functionalities, further reinforcing the integration with Santa’s technological stack. In particular, we are working to make our solution compatible with the ML-enabled decision support system that augments Santa’s capacity to make a list, check it twice, and find out who’s naughty or nice. In this way, the gift delivery route will be modified in real time based on last-minute reconfigurations of the list.

If you want to learn more about Helicon and the way in which it can help your business enhance decision making with artificial intelligence and streaming data, visit our website and start your free demo.

In the meantime, the Radicalbit team wishes you a very special Christmas and a wonderful new year!