APPROACH
The approach to address the client’s challenge included:
- Building re-usable data pipelines to move away from the current Hadoop-based system to GCP
- Leveraging the GCP environment to build a cost-optimized ecosystem to cater to both the business users and the data-scientist community
- Merging demographics, customer, geo-location, credit-model, clickstream & marketing campaign data to create a single source of truth
Using a model management on GCP to track and maintain 150+ ML models
KEY BENEFITS
Our solution helped the client:
- Develop an auto-scalable infrastructure on GCP to manage variable workloads thus reducing cost
- Use big-query and data-proc in tandem to provide compute-horsepower on a case-to-case basis based on cost
- Leverage Kubeflow and Kubernetes for model management and deploying model endpoints for down-stream consumption
RESULTS
- The UDP since its inception has been processing 250 TBs (Terabytes) of data weekly
- Overall reduction of processing time in computation-intensive jobs by 70 percent
- Overall costs reduced by 30-35 percent