Skip to content

Latest commit

 

History

History
99 lines (65 loc) · 5.68 KB

README.md

File metadata and controls

99 lines (65 loc) · 5.68 KB

Foundry Local

Livestream starting soon! Click the image below to watch the recording.

Reactor Livestream

Building AI Agents E2E On Microsoft Foundry

In this Quest, you'll explore the end-to-end model development journey for designing and developing AI agents with Microsoft Foundry models. You'll start at the Foundry portal and learn to setup a project, then select and deploy the right model for the task. Then, you'll move to code, and learn to observe, optimize, and protect, your AI agent using techniques like fine-tuning, tracing, evaluations, and AI red-teaming - using the Microsoft Foundry SDK. By the end of this quest, you should have a good intuition for the capabilities of the Microsoft Foundry platform and have a sandbox you can use to explore these further.

Scenario

Let's use a popular real-world scenario to motivate the quest. Imagine you are an AI engineer working for Zava, a fictitious enterprise retail company. You have been asked to build Cora, your new customer service AI agent that answers shoppers' questions about products in this sample catalog.

Cora needs to meet three requirements:

  1. Be polite and helpful in interactions. Think tone and response format.
  2. Be cost-effective to operate. Think latency, compute and token costs.
  3. Be trustworthy in responses. Ensure safety, quality and accuracy.

How do we go from plan to prototype and production, to meet these goals?

Steps To Complete The Quest

Note on costs associated with the quest

1. Pre-Requisites

To complete the quest, you will need:

  1. an active Azure subscription - with Owner or Contributor access.
  2. a personal GitHub account - with Codespaces quota.

2. Launch Codespaces

  1. Open a new browser tab and log into your GitHub account.
  2. Now navigate to this link to see the page below. Click create codespace. Codespaces
  3. This will open up a new browser tab with a VS Code editor session loading.
  4. Wait till the VS Code editor has loaded - and you see an active terminal. This takes a few minutes.
  5. Your local environment is ready. ✅

3. Authenticate with Azure

  1. Type this command into the VS Code terminal when ready. Then complete the prompts using the device code, to log into your Azure account. 
    az login
  2. Verify that you are logged in using this command. You should see your Azure profile. 
    az account show
  3. Your local environment is connected to Azure. ✅

4. Complete the Quest

Note

The Foundry SDK is under active development.

If you encounter any blockers, unclear steps, or have suggestions to improve the developer experience, please open an issue using the provided issue template - Bug report/ Build-a-thon Feedback. Make sure to apply the label that matches the step you were on when the issue occurred.

Buildathon issue with labels

The repoistory is setup with a devcontainer.json that installs all the required depdencencies and gives you access to a free tier of GitHub Copilot.

Open the README.md in that repository to see the various tasks available for that quest.

  • Task 1: Understand Foundry capabilities
  • Task 2: Setup a Foundry project
  • Task 3: Select a base model
  • Task 4: Customize the base model
  • Task 5: Design the AI Agent
  • Task 6: Evaluate the agent responses
  • Task 7: Trace the agent execution
  • Task 8: Run a Red-Teaming scan
  • Task 9: Teardown and cleanup

Important Notes:

  1. Some tasks like fine-tuning can take a lot of time to complete - in this case, you should be able to go on to the next task, then return to review results here when the job completes.
  2. Tasks like fine-tuning, evaluations and red-teaming have added costs and constraints - for instance, we need to use a specific model in a specific region. If this is not viable in your subscription, then treat this as a read-only task and explore the code and results from our run.
  3. Teardown the project when done - to prevent unexpected costs from model or compute use.

This quest has a lot of steps to complete, many of which cannot be completed in a single hour. And that's okay. The main objective is to give you intuition for what an end-to-end journey would involve - and provide code snippets and results from a sample run, to help you connect theory to practice.

Focus on completing just a few tasks today. Validate your development envrionment and get familiar with the Microsoft Foundry UI and SDK. Then, read through the rest using the screenshots from our sample execution run to get an understanding of the purpose and process for each step.

Explore the rest at your own pace later. Create a fork of the repo, then launch a new Codespaces instance on that fork and work through the steps in order. Try customizing the data to suit a different scenario. Or change the code to try a different evaluator or attack strategy for red-teaming. Use your fork as a sandbox for building a deeper understanding of these capabilities with hands-on experiments.

Return to the Build-a-thon

Once you have completed this quest and get an intuitive sense for end-to-end development with Microsoft Foundry, return to the main Build-a-thon repository to continue with the next quests.

AI Note

This quest was partially created with the help of AI. The author reviewed and revised the content to ensure accuracy and quality.