Safe, Verifiable, & Maintainable AI Apps in Python

SciPy 2026 Tutorial

Carson Sievert · Posit

A 4-hour hands-on tutorial on building safe, verifiable, and maintainable AI apps in Python — using chatlas, shinychat, and querychat.

Slides

  1. Welcome & Setup
  2. Foundations
  3. Intro to chatting with LLMs (chatlas)
  4. Recap
  5. Tool calling and agents
  6. Structured data extraction
  7. Chatbots
  8. Recap
  9. querychat
  10. Evals
  11. Observability (OTel)
  12. Recap & Sendoff

Setup

A computing environment with all of the dependencies needed to run exercises is available here. Please follow the instructions in the welcome slides to get set up.

Local Setup

If you’d rather locally run the tutorial exercises, follow the following instructions to get set up.

  1. Clone the repo
git clone https://github.com/cpsievert/scipy26-tutorial.git
cd scipy26-tutorial
  1. Install uv globally
# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

If you can’t use uv for some reason, Just do what you’d normally do to install the requirements.txt

  1. Setup a virtual environment
uv venv
source .venv/bin/activate
  1. Install dependencies
uv pip install -r requirements.txt

LLM access

Exercises depend on access to an LLM. Here are a few different ways to get access:

  1. Sign up for a free trial at https://posit.ai/. No credit card required. You can use the free trial for the entire workshop.

  2. Bring your own API key. If you have your own API key, like OPENAI_API_KEY, ANTHROPIC_API_KEY, etc. that will work just fine.

  3. Run a local model on your laptop via LM Studio or Ollama. No key required, fully private, but slower and requires a beefy laptop. See the LM Studio docs for setup instructions.

For more choices and details, see https://posit-dev.github.io/chatlas/get-started/models.html

Exercises and more

Can be found in the GitHub repo. See the exercises/ folder.

Resources