Building Your First Workflow

Lesson Overview

In this video, you'll learn the fundamentals of building your first workflow in AirOps. The lesson covers essential concepts like setting up inputs, using Google Search for context gathering, crafting effective LLM prompts, and testing your workflow.

• 0:00: Introduction to building your first workflow
• 0:23: Overview of the AirOps studio interface
• 2:10: Setting up inputs and gathering context with Google Search
• 5:20: Crafting effective LLM prompts and choosing the right model
• 11:39: Testing your workflow and using the grid feature

Key Concepts

AirOps Studio Interface

The AirOps studio is where you construct and test your workflows. The canvas on the right is where you drag and drop different blocks, configure them, and add notes. The left-hand side contains the various steps and blocks you can use in your workflow.

Liquid Templating Language

Liquid is a templating language used in AirOps for referencing variables and formatting text. It's important to have a basic understanding of Liquid syntax when building workflows, as it allows you to pass runtime variables and accumulate context throughout the workflow.

JSON Data Structures

JSON (JavaScript Object Notation) is a lightweight data interchange format used in AirOps workflows. Familiarizing yourself with JSON is crucial for understanding the structure of data returned by various steps, such as Google Search results.

Crafting Effective LLM Prompts

When creating an LLM prompt, the model choice, system prompt, and chat message are the most important factors. The system prompt sets the context and role for the model, while the chat message contains the instructions and references to variables. Providing example outputs in the prompt can help ensure consistency in the model's responses.

Key Takeaways

1. Start by defining the inputs your workflow requires and setting up the necessary input fields.
2. Use the Google Search step to gather relevant context for your LLM prompts.
3. Choose an appropriate model based on the task complexity and your performance requirements. Popular choices include Claude Sonnet, GPT-4, and 01 for more complex tasks.
4. Craft clear and concise LLM prompts, including a well-defined system prompt and chat message with instructions and variable references.
5. Provide example outputs in your prompts to ensure consistency in the model's responses.
6. Test your workflow using the grid feature to run multiple iterations and review the results.

By understanding these key concepts and following the steps outlined in the video, you'll be well on your way to building powerful and effective workflows in AirOps.