Run Positron on Amazon SageMaker AI for data science workflows
Positron, Posit’s integrated IDE for data science, is now available as a custom container on Amazon SageMaker Studio, enabling data scientists to access Athena, Glue Data Catalog, and S3 without managing credentials and to run AI‑assisted coding via Posit Assistant backed by Amazon Bedrock. The workflow demonstrates a synthetic 50,000‑loan portfolio where Athena queries feed R feature validation and Python‑based XGBoost training, with the model deployed to a real‑time endpoint and accessed from Shiny for Python and Quarto. Engineers can provision the Positron image in Amazon ECR, register it with SageMaker Studio, and launch Spaces on ml.t3.xlarge (or larger) instances, gaining governed data access and in‑account LLM assistance.
⚡ Key Takeaways
- Synthetic 50,000‑loan portfolio used for the end‑to‑end demo.
- Positron runs from a custom container built on the Amazon SageMaker Distribution image and registered to a Studio domain.
- Compute is provisioned on ml.t3.xlarge (or larger) instances, with capacity reservable via SageMaker AI training plans.
- Posit Assistant leverages Amazon Bedrock models within the same AWS Region for AI coding help.
- Administrators must push the Positron image to an Amazon ECR repository and configure a Space execution role with Athena and Glue permissions.
By embedding a full‑stack data‑science IDE directly in SageMaker Studio, teams eliminate credential sprawl, streamline model training and deployment pipelines, and keep LLM‑driven assistance inside their trusted AWS environment.
✅ Practical Steps
- Build the Positron container from the Posit‑published definition, push it to your Amazon ECR repository, and register it with SageMaker Studio.
- Attach the custom image to a Studio domain, create a Space using the ml.t3.xlarge (or larger) instance type, and select Positron as the IDE.
- Enable Posit Assistant by granting the Space execution role access to Amazon Bedrock models in the same AWS Region.
Want the full story? Read the original article.
Read on AWS ML Blog ↗