Implement dummy data utility

[bfollington]

Introduces a small utility function to trivially generate data based off of a description, some input data and a schema it should adhere to. Should be useful for implementing a prototype of search (cc @seefeldb).

import { generateData, LLMClient } from "@commontools/llm-client";

const client = new LLMClient({
  serverUrl: LLM_SERVER_URL,
  tools: [],
  system: `Generate dummy data as JSON as per the provided spec. Use the input to imagine what an API response would look like for a request.`,
});

const test = await generateData(
  client,
  "generate 3 user profiles",
  { name: "John Doe" },
  {
    type: "array",
    items: {
      type: "object",
      properties: {
        name: { type: "string" },
        email: { type: "string" },
        age: { type: "number" },
        role: { type: "string" },
        birthday: { type: "string" },
        favoriteColor: { type: "string" },
      },
    },
  },
);

console.log(test);

This produces:

[
    {
        "name": "John Doe",
        "email": "john.doe@example.com",
        "age": 35,
        "role": "Software Engineer",
        "birthday": "1988-03-15",
        "favoriteColor": "blue"
    },
    {
        "name": "Jane Smith",
        "email": "jane.smith@example.com",
        "age": 28,
        "role": "Product Manager",
        "birthday": "1995-07-22",
        "favoriteColor": "purple"
    },
    {
        "name": "Alex Johnson",
        "email": "alex.johnson@example.com",
        "age": 42,
        "role": "Marketing Director",
        "birthday": "1981-11-03",
        "favoriteColor": "green"
    }
]