Get data science off the desktop
When your data lives in Snowflake but your analytics stack lives elsewhere, workflows grind to a halt. Teams end up juggling fragmented local samples or waiting on IT infrastructure requests, creating security blind spots and slowing time-to-insight.
This guide shows data scientists and data science leaders how to run R and Python against Snowflake data, with AI assistance from Snowflake Cortex, without moving the data.
What's inside
- R and Python on Snowflake data: open source packages let you query Snowflake from R and score models inside the warehouse.
- AI assistance on Snowflake Cortex: Posit Assistant in Positron uses Cortex as its LLM backend, with Claude and GPT models and no API key to request or rotate.
- AI suggestions a reviewer can verify: Posit Assistant grounds each suggestion in your code, data, and environment, and reads your Snowflake semantic model.
- Shipping without an engineering handoff: publish from your IDE to Posit Connect in one click, as a Shiny or Streamlit app or a scheduled report.
- Proof from teams like yours: Brilliant Earth ships in hours what previously took weeks, and Pinterest analyzed 30,000 survey comments that SQL alone could not.
Who is this guide for
This guide is for data scientists and data science leaders who work in R and Python and keep their data in Snowflake. If you’re painfully pulling samples to a laptop, waiting on IT for a VM, or give finished models to an engineering team to deploy, start here.
Frequently asked questions
Can I use an AI coding assistant on Snowflake data without a separate LLM API key?
Yes. When you enable Posit Assistant in Positron, it configures Snowflake Cortex as its LLM backend and uses the authorization you already have, so you have no API key to request or rotate. Your Snowflake role must have Cortex models enabled. Cortex also gives Posit Assistant access to Claude and GPT models, so you pick the model for the task.
How does Posit Assistant help a reviewer trust AI-generated code?
Posit Assistant grounds each suggestion in the code, data, and environment behind it, so a reviewer can rerun the result and confirm it. When your team has defined a Snowflake semantic model, Assistant reads it, so suggested joins and field names match your team's definitions instead of a guess. It also checks in before steps that are expensive or risky.
How do I get a finished model or dashboard to stakeholders without an engineering team?
Publish from your IDE to Posit Connect in one click as a Shiny or Streamlit app, a Quarto report, an API, or a scheduled job. On the Native App, Connect runs it inside Snowflake, and each viewer sees the data their own Snowflake role allows.