Event date is -
Governed AI for data science with Posit and Amazon Bedrock
Governed AI for data science means your team can use AI assistants and agents while IT decides which models you can use, what data you can send, how AWS logs each request and what it costs. Amazon Bedrock holds those controls in one place, and Posit tools use them.
In this demo video, James Blair (Posit) and Rodrigo Gazzaneo (AWS), discuss how data scientists use Posit Assistant in Positron on Amazon SageMaker while guardrails, budgets and usage reporting stay in place.
What you'll see:
- How Amazon Bedrock keeps prompts and data inside AWS and lets you choose the regions that process your requests
- Posit Assistant in Positron on Amazon SageMaker, using your existing AWS permissions with no API keys to manage
- Model access by user, guardrails that block PII and policy violations, and the cost of every Posit Assistant turn
- Exploring a loan dataset in Amazon Athena with Posit Assistant, and tracking Bedrock usage by user, model and tokens
- An agent built with Commons that lets business users ask questions of data in plain language, and flags answers based on trusted calculations
- Posit Assistant also runs in RStudio and Posit Workbench, with the same Bedrock controls.
Key takeaways:
- Amazon Bedrock gives IT one place to control model access, guardrails, logging and budgets for AI used in data science.
- Posit Assistant in Positron, RStudio and Posit Workbench uses Bedrock through existing AWS permissions, with no API keys.
- Guardrails apply at the model, so the same rules hold in every tool that calls it.
- Commons lets business users ask questions of data in plain language and see which answers come from trusted calculations.
FAQ:
What is governed AI for data science?
It is using AI assistants and agents in data science work under organizational controls: which models each person can use, what data can be sent, how every request is logged and how much it costs. In this demo, Amazon Bedrock holds those controls and Posit tools use them.
How does Amazon Bedrock keep data private?
AWS hosts the models in Bedrock, so prompts, responses and context stay within AWS. Bedrock does not use prompts and responses to train models and does not share them with model providers. US-only inference profiles keep processing in US Regions, and AWS PrivateLink keeps network traffic private.
Who can use AI, and which models?
Only the people and roles you allow can use AI, and only with the models you approve. You set this with IAM policies in AWS, not in the tool. If someone picks an unapproved model in Posit Assistant, Bedrock returns an access-denied error.
Can Posit Assistant use models from Amazon Bedrock?
Yes. Posit Assistant in Positron and RStudio can use Bedrock models. On Amazon SageMaker it uses the space execution role, and on Posit Workbench it uses Workbench-managed AWS credentials, so you have no API keys to manage.
How do Amazon Bedrock guardrails work with Posit Assistant?
Admins set guardrails on the model in Bedrock, so they apply to any tool that sends a request. In the demo, Bedrock blocked a prompt that contained a Social Security number and a request to decline loan applicants by age. Posit Assistant then ended the conversation.
How can teams control AI costs with Amazon Bedrock?
Admins can limit which models each user can access, track spend by IAM role in CloudWatch and Cost Explorer, and set budget alerts. Posit Assistant also shows you the cost of each session and each turn.
What is Commons?
Commons is an R and Python package that helps data scientists build trustworthy data analysis agents that reach for your team's trusted code first, cite trusted documentation when they can't, and clearly flag it when they have to improvise, so every answer comes with proof of how it got there.
How do I get started with Positron on Amazon SageMaker?
Get a Posit license in AWS License Manager, either through the Posit Team listing on AWS Marketplace or directly from Posit. An admin then sets up the custom image in your SageMaker domain, and users can create Positron spaces.
Does saving prompts in a Quarto file make AI results reproducible?
No. AI assistants are non-deterministic, so the same prompt can return different output. Tools that record prompts and agent turns alongside your analysis help you see how you reached a result.
Resources:
Speakers