Your web browser is out of date. Update your browser for more security, speed and the best experience on this site.

Update your browser
CapTech Home Page

Summary

CapTech collaborated with a Top 5 Quick Serve Restaurant (QSR) to expand the capabilities of its analytics platform used by dozens of data scientists and engineers. The new Machine Learning (ML) Ops platform now facilitates the development of contemporary data pipelines, workflow orchestration, and ML model serving for several enterprise verticals including, supply chain, accounting, and operator management.

Challenge

The fast food restaurant asked CapTech to augment its capabilities in ML engineering in response to the growth of the platform and its digital infrastructure. The goal of the project was to decrease development time for engineers and provide an intuitive and easy experience for data scientists to both test and deploy their models. These models serve all parts of the business, and improving processes around deployment and iteration was vital to expanding the use cases of ML at the enterprise.


In addition to serving internal customers of the platform, the improved usability and expanded feature set has allowed the platform to incorporate more kinds of models, including some targeted at external customers.

Approach

Captech has contributed to several major products and releases throughout the project. Our consultants were responsible for building new features for the internally developed Python package that allows users to easily connect to and use AWS resources from Airflow. The use of this package contributes to the unique MLOps experience at the client and has resulted in a dramatic reduction in development time for new models. Through this development, CapTech was entrusted with the responsibility of supporting data scientists in productionizing and deploying ML models they had developed.


In addition to supporting the existing platform, CapTech consultants performed an evaluation of Databricks as an alternative development platform for MLOps. As a result of this evaluation, Databricks became the new standard for the company and CapTech has been responsible for testing and building new best practices, as well as onboarding data scientists to the new platform. Use of Databricks has generated many improvements including, improved scalability, direct integration of ML Tools, easier permissions sharing, and a notebook development environment for the analytics team.

Results

The development of this platform has facilitated a massive expansion in the fast food restaurant’s ML footprint. In the year that CapTech has supported this platform, it has onboarded 60+ new ML pipelines that now run in production. The improved usability and direct integration of tools has also improved time to production, reducing engineering overhead and allowing a much more self-service platform than previously.


The model supports many functions across the business and many internal and external customers. Internally, the platform supports models such as:

  • Customer lifetime value: Consumption habits of customers at restaurants and predictions on total lifetime spend and transaction volumes.
  • Automated accounting review: Anomaly detection algorithm that flags reimbursement requests from operators that are not legitimate.


Externally, the MLOps platform provides models like:

  • Estimated wait times: Real-time model for the online ordering application, which predicts time until an order is ready.​
  • Item recommendations: A real-time recommender system for the customer-facing loyalty application, which provides recommendations based on purchase patterns.