GS Caltex accelerated ML model deployment by connecting Amazon SageMaker Canvas and Pipelines
GS Caltex connected no-code model development in Amazon SageMaker Canvas with production deployment in Amazon SageMaker Pipelines. According to the company, deployment time was shortened from a week to hours, and batch inference every 30 minutes reduced operating costs.
The South Korean energy and chemical company GS Caltex has introduced an MLOps workflow that connects process engineers who build and validate models without writing code in Amazon SageMaker Canvas with ML engineers who configure deployment and operations in Amazon SageMaker Pipelines and Amazon SageMaker Model Registry. The company created an automated handoff process that reuses artifacts (preprocessing procedures and trained models) created in SageMaker Canvas without having to manually rewrite them for the production environment.
According to GS Caltex, this has shortened the time needed to deploy a new model from approximately one week to a few hours. Batch inference now runs every 30 minutes instead of operating continuously running endpoints, which, according to the company, has significantly reduced inference costs; the source does not provide a specific figure for the savings.
The workflow divides the work according to expertise: process engineers understand the variables and data of the given production process and assess the usefulness of the model's predictions, while ML engineers take over the validated artifacts, approve model versions in Model Registry, and manage their runs, run history, and error handling. Run scheduling is handled by Amazon EventBridge, input and output data are stored in Amazon S3, and step status can be monitored via Amazon CloudWatch.
The source article further describes the technical setup steps (exporting data via SageMaker Processing, managing .flow files, checking container images and output nodes), but this part is missing entirely from the source. See the source article for details.
Why it matters
It shows a concrete pattern of dividing work between domain experts without programming knowledge and ML engineers, which shortens the time from creating a model to its production deployment and replaces continuously running compute resources with batch processing. This is relevant for companies that operate prediction models tied to operational processes and want to reduce both infrastructure costs and team time.
Two audiences, two different impacts
What this means
For individuals
ML and data engineers can use this pattern (no-code prototyping by a domain expert + standardized handoff into production pipelines) to shorten the time needed to deploy their own models.
For a business
According to this case, companies with process or domain experts can shorten ML model deployment from weeks to hours and reduce inference costs by switching to batch processing instead of continuously running endpoints.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.