MLOps case study: Increasing profits with MLOps

MLOps for Business

Arne Jonas Warnke

Head of Emerging Curriculum

Real-life case study

$$

$$

Case study

$$

  • Real-life
  • To bring all together
MLOps for Business

Case study: cooling water demand and availability

$$

Business question:

Predict amount of cooling water that is required and available in the next two weeks

Being prepared for potential prop bottlenecks.

$$

MLOps for Business

Modeling cooling water

$$

Predicting cooling water over

  • on an hourly base
  • the next two weeks

$$

24 hours * 14 days = 336 forecasts (each hour)

$$

Presented in dashboard

  • to management and engineers
MLOps for Business

Modeling cooling water

$$

Forecast based on

  • Internal data
    • Sensors
    • Production planning

$$

  • External data
    • Weather forecast

$$

MLOps for Business

The team

$$

Team composition:

  • Data scientist
  • Data engineer
  • Data architect
  • Backend engineer (occasional)

No software engineer

MLOps for Business

Collaboration

$$

$$

Collaboration

$$

  • No separation between development and operations
  • Autonomous
  • Fast to respond to unforeseen events
MLOps for Business

Project progression

  1. Initial weeks
    • Clear business objective
    • Data pipelines
    • Baseline model

$$

  1. Medium-term
    • Better data
    • Better model

$$

MLOps for Business

Project progression - infrastructure

  1. Initial weeks
    • Clear business objective
    • Data pipelines
    • Baseline model
    • Preliminary infrastructure

$$

  1. Medium-term
    • Better data
    • Better model
    • Better infrastructure

$$

Logo GitLab

MLOps for Business

Conclusion

$$

Project was a success

  • Business objectives met
  • Valuable information in critical times

$$

But

  • few downtimes
  • failed to streamline the application

$$

MLOps for Business

Our MLOps maturity

$$

Level Description Highlights
0 No MLOps Siloed teams, manual processes
1 DevOps no MLOps Siloed teams, automatic data gathering, first automatic tests
2 Automated Training Better collaboration, reproducibility, easier deployment
3 Automated Deployment Good collaboration, full reproducibility, traceability, automated testing
4 Automated Re-training Full collaboration, mastering operations, very few downtimes
1 https://learn.microsoft.com/en-us/azure/architecture/example-scenario/mlops/mlops-maturity-model
MLOps for Business

Let's practice!

MLOps for Business

Preparing Video For Download...