AI Production System: Data-Science Platform Modernization

The leading US online mortgage lender engaged Taller to provide senior cloud-infrastructure engineering as it migrated its data-science operation from aging on-premise servers to AWS. Taller’s engineers worked across the stack the lender’s machine-learning models ran on.

The lender’s mortgage decisions ran on a data-science operation: models that scored risk, predicted outcomes, and increasingly made calls automatically. But that operation was running on aging in-house servers that couldn’t keep up with the demand being placed on them. The lender set out to move the whole thing to the cloud (using Amazon’s AWS services for storing data, crunching it, and running its machine-learning models), with a careful setup that walled off the live production system so that teams experimenting with new models could look but couldn’t accidentally break anything. Pulling off a migration at this scale, without disrupting a system the business depended on daily, took serious cloud-engineering muscle.

Taller supplied lead- and senior-level cloud-infrastructure engineers to the lender, working primarily across its AWS stack: SageMaker for building and training models, EMR and Glue for large-scale data processing, S3 and Redshift for storage and warehousing, Athena and Lake Formation for querying and governing the data lake, Lambda for serverless compute, and EKS with Istio running the online model-scoring services.