​

Data Transformation

Automate conversion pipelines by having an LLM produce transformation functions, then execute safely via AVM.

​

Use Cases

​

Web2: CSV → JSON → Notion

Transform business data across different formats and integrate with productivity tools.

​

Web3: Normalize Wallet Data → Subgraphs

Process blockchain wallet data and prepare it for indexing in decentralized subgraphs.

​

Scenario: Custom Schema Delivery

Clients request data in bespoke JSON structures without manual coding.

​

Implementation: Schema-Driven Codegen

1. Define Schema
   Provide a JSON Schema template.
2. Generate Code
   Prompt the LLM to write an execute(input) function matching the schema.
3. Secure Run
   Execute the function in AVM’s sandbox.
4. Deliver Output
   Return parsed JSON conforming to the schema.

​

Example (TypeScript)

import { runPythonTool } from "@avm-ai/avm-vercel-ai";
import { generateText } from "ai";
import { openai } from "@ai-sdk/openai";

const tool = runPythonTool();

async function transform(sample: string, schema: object) {
  const prompt = `
Write execute(input) that transforms CSV in input["data"] to JSON
following this schema: ${JSON.stringify(schema)}`;
  const { text: code } = await generateText({
    model: openai("gpt-4o"),
    prompt,
    tools: { runPython: tool },
  });
  const result = await tool.exec({ code, input: { data: sample } });
  return JSON.parse(result.output.result);
}

​

Next Steps

• Validate output with JSON Schema tooling.
• Chain pipelines in agent orchestrations.
• Support additional runtimes for performance-critical tasks.