ML-based demand forecasting solution to improve forecast accuracy while reducing costs by 5x

Sigmoid Solution

Sigmoid developed an agile demand forecasting solution by pulling data from Azure Data Lake storage, preprocessing it to remove outliers and running feature engineering on it. We trained several ML and deep learning models that ran parallely across all countries on Azure Kubernetes Service (AKS) to generate a trained model representing the behavioral aspects of the data to determine specific patterns. The trained model was used to create forecasts that were closely monitored through Grafana. To modularize the code, Sigmoid built the entire framework on Kedro that offers best practices for software engineering and makes the code easy to scale. In addition, we operationalized the final solution by parameterizing the end-to-end pipeline countrywise with auto-enabling features on AKS that scale up and down based upon the number of countries in the run.