Customer
The Tier-2 Eastern European bank is a full-service institution with an impressive AUM of over $50 billion, offering a complete portfolio of banking services for both private individuals and companies.
Challenge
Long time-to-market for new and modified credit risk models. High operational risk and efforts associated with deployment of ML-models in production environments and data streams. Poor model performance monitoring in production.
Project Results
The new MLOps Platform enabled quick implementation of ML-models of any complexity into business processes and eliminated operational risks associated with deployment of models into production environments. The deployed platform architecture automates model management, experiment tracking, code and data versioning as well as model validation, deployment and serving.
Business Value
Dramatic reduction of time-to-market for new ML-models. Improved model-driven services performance thanks to the new state-of-the-art deployment architecture running on Kubernetes cluster. Simplified onboarding of new data science team members, now taking less than 1 day.
Tech Stack
JupyterHub, Tensorflow, Scikit-learn, MLFlow, Apache Airflow, Apache Spark, Cloudera CDP, Kubernetes, Jenkins, ArgoCD.
The Tier-2 Eastern European bank is a full-service institution with an impressive AUM of over $50 billion, offering a complete portfolio of banking services for both private individuals and companies.
Challenge
Long time-to-market for new and modified credit risk models. High operational risk and efforts associated with deployment of ML-models in production environments and data streams. Poor model performance monitoring in production.
Project Results
The new MLOps Platform enabled quick implementation of ML-models of any complexity into business processes and eliminated operational risks associated with deployment of models into production environments. The deployed platform architecture automates model management, experiment tracking, code and data versioning as well as model validation, deployment and serving.
Business Value
Dramatic reduction of time-to-market for new ML-models. Improved model-driven services performance thanks to the new state-of-the-art deployment architecture running on Kubernetes cluster. Simplified onboarding of new data science team members, now taking less than 1 day.
Tech Stack
JupyterHub, Tensorflow, Scikit-learn, MLFlow, Apache Airflow, Apache Spark, Cloudera CDP, Kubernetes, Jenkins, ArgoCD.