Basic Extraction API¶

With Marvin, you can easily build a production-grade application or data pipeline to extract structured data from unstructured text. In this example, we'll show how to

Write an AI-powered Function.
Build a production-ready API.

Writing an AI-powered Function.¶

Example

Marvin translates your Python code into English, passes that to an Large Language Model, and parses its response. It uses AI to evaluate your function, no code required.

Let's build a function that extracts a person's first_name, last_name, and age from a paragraph of text. This is typically a pretty daunting task for a machine learning savant, much less your average engineer. Marvin lets you accomplish this by write code the way you normally would, no PhD required.

We'll simply write a Python function, tell it that we expect a text input and that it'll output a dict, or dictionary. With Marvin, we'll use ai_fn and decorate his function. When we do, this function will use AI to get its answer.

from marvin import ai_fn, settings
from typing import Any

settings.openai.api_key = 'API_KEY' 

@ai_fn
def extract_person(text: str) -> dict:
    '''
        Correctly infers a persons `birth_year`, `first_name` and `last_name`
        from the passed `text`. 
    '''

This function can now be run! When we test it out, we get great results.

Results

extract_person('My name is Peter Parker, and I was born when Clinton was first elected')

# returns {'first_name': 'Peter', 'last_name': 'Parker', 'birth_year': 1992}

Build a production-ready API.¶

In the following example, we will demonstrate how to deploy the AI function we just created as an API using FastAPI. FastAPI is a powerful tool that allows us to easily turn our AI function into a fully-fledged API. This API can then be used by anyone to send a POST request to our /extract_person/ endpoint and get the structured data they need. Let's see how this can be done.

Example

Now that we have our AI function, let's deploy it as an API using FastAPI. FastAPI is a modern, fast (high-performance), web framework for building APIs.

from fastapi import FastAPI
from marvin import ai_fn, settings

app = FastAPI()

settings.openai.api_key = 'API_KEY'

@app.post("/extract_person/")
@ai_fn
def extract_person(text: str) -> dict:
    '''
        Correctly infers a persons `birth_year`, `first_name` and `last_name`
        from the passed `text`. 
    '''

With just a few lines of code, we've turned our AI function into a fully-fledged API. Now, anyone can send a POST request to our /extract_person/ endpoint and get the structured data they need.

API Deployment

import uvicorn
import asyncio

config = uvicorn.Config(app)
server = uvicorn.Server(config)
asyncio.run(server.serve())

Now, you can navigate to localhost:8000/docs to interact with your API.

Making Requests

import requests

data = {"text": "My name is Peter Parker, and I was born when Clinton was first elected"}
response = requests.post("http://localhost:8000/extract_person/", json=data)
print(response.json())

# returns {'first_name': 'Peter', 'last_name': 'Parker', 'birth_year': 1992}

This will send a POST request to the /extract_person/ endpoint with the provided text and print the response.