Key Takeaways

Recently, I’ve been tinkering on a project leveraging the Vercel SDK to test out this cool new hype term called AI Agents. Matt Pocock over at Total TypeScript has an amazing course that walks you through examples of using the SDK effectively. He also walks you through effective Prompt Engineering right from the command line and not using a UI directly seems to be a fast experience using the OpenAI and Anthropic models. Let’s walk through some of my highlights.

Structured Outputs

What you’re seeing is a basic structured input/output to create a recipe (which I myself as a health fanatic hate making my own meal prep).

import { generateObject } from "ai";
import { z } from "zod";
import { openai } from "@ai-sdk/openai";

const model = openai('gpt-4');
const schema = z.object({
    recipe: z.object({
        name: z.string(),
        ingredients: z.array(z.object({
            name: z.string(),
            amount: z.string()
        }).describe('The ingredients needed for the recipe.')
    ),
    steps: z.array(z.string()).describe('The steps to make the recipe.')
    }),
});

export const createRecipe = async (input: string) => 
{
    const { object } = await generateObject({
        model,
        schema,
        schemaName: 'Recipe', 
        prompt: input,
        system: `
        You are helping a user create a recipe. 
        Use British english variants of ingredients names, like Coriander over Cilantro.
`
    });

    return object.recipe;
}

const recipe = await createRecipe(`Create a recipe for a simple pasta dish with tomatoes and basil.`);

console.log(recipe);

Use Cases

I can see a use case where I want to develop my own recipes daily or weekly and AI could help me make a simple menu for the week. Might try to set up a timer in ChatGPT and see if it makes a difference vs. building out my own.

Check out the GitHub with the code for examples: https://github.com/rymanacevedo/vercel_ai_sdk