Skip to content

3. Ideate With Prompty

THIS SECTION IS FOR INFORMATION ONLY → Skip to the next section to get started on exercises.

The infrastructure is provisioned. The development environment is setup. It's time to ideate.

Dev Workflow

3.1 What are we building?

The Contoso Chat sample has a working prototype of the chat AI in the src/contoso-chat/api/contoso_chat folder. These are the components of that application. 

contoso-chat/
    product/
        product.prompty
        product.py
    chat.json
    chat.prompty
    chat_request.py
In this section, we'll recreate the application from scratch using the process of ideation.

3.2 What is ideation?

Ideation is the process of going from the first prompt, to a functioning prototype. To do this:

  • We start with a test question (user prompt) and an LLM (chat model)
  • We assess the model's response to the question to see if meets a desired quality.
  • If not, we "engineer" the prompt or "configure" the model, and try again.

In this lab, we'll mimic iteration by starting with a basic.prompty asset, then renaming it to a new asset of the form chat-n.prompty to reflect the n-th round of iteration. The goal is to get to the same chat.prompty version seein in the finished application above.

3.3 What is Prompty?

We talked about chat.prompty as our application prototype. What exactly is a .prompty file?

Prompty is an open-source project from Microsoft that defines a specification (.prompty file format) and provides tooling (to create assets) and runtimes (to execute assets). Think of Prompty as a technology that brings the model playground experience to your Visual Studio Code editor!

With Prompty you can:

  • Define the prompt template in a .prompty file using YAML syntax
  • Configure app metadata and model configuration paramters in frontmatter
  • Specify input types (with sample data) that map to template context
  • Execute the prompt with a model invocation, right from the editor
  • Convert the asset into executable code in a supported language (e.g, Python)

This lets us iterate rapidly within our development environment, to shape the data for a RAG-based application and evaluate it for quality before deploying to production. Let's see this in action!