← Casesbank
An on-prem ML platform on ClearML
Consulting and custom development for a bank: an ML platform on the bank's own servers and a real business task on top of it. The details are under NDA, so this case is short.
Numbers
on-premplatform on the bank's servers
4open source components: ClearML, Feast, JupyterHub, KServe
Helm → prodthe platform comes up from code, not by hand
Before
- The ML platform had to run on the bank's servers so that the bank could bring it up again on its own.
- The customer churn model needed a path from training to inference with no manual steps.
After
- The ML platform is deployed on-prem with Helm charts, and the bank can bring it up again on its own.
- The customer churn model is trained and rolled out to inference automatically.
What we did
Handed the bank IaC for the ML platform on open source: ClearML, Feast, JupyterHub and KServe. The platform is deployed with Helm charts on the bank's servers, from code, not by hand.
Solved a real bank task on the platform: customer churn. Built a pipeline that trains the model and rolls it out to inference with no manual steps.
Stack
ClearMLFeastJupyterHubKServeKubernetesHelm
My role
Consulting and development: IaC for the platform on Helm and the churn model pipeline.