Skip to main content

Building the API

Updated May 14, 2023 ·

Exposing the Model​

To make our model accessible, we expose it through an API (Application Programming Interface). APIs let applications communicate, just like user interfaces let humans interact with software.

  • Server – Hosts the model and handles requests
  • Client – Sends data to the server and receives predictions
  • API Types – REST, RPC, SOAP, and others

APIs allow clients to run procedures (like sentiment analysis) or access remote data (like news articles).

API Example​

Below is a sample API for a navigation app that estimates arrival time.

  • Takes velocity and distance as input
  • Runs them through the model
  • Returns estimated arrival time

A simple API using FastAPI in Python:

from fastapi import FastAPI

app = FastAPI()

@app.get("/estimate")
def estimate_time(velocity: float, distance: float):
    if velocity <= 0 or distance <= 0:
        return {"error": "Velocity and distance must be positive numbers"}
    time = distance / velocity
    return {"estimated_time": round(time, 2)}

Example request:

curl "http://127.0.0.1:8000/estimate?velocity=60&distance=120"

Expected response:

{"estimated_time": 2.0}

Input Validation​

To prevent errors, we validate input data before processing it.

  • Ensure required fields – Check for missing inputs
  • Verify data types – Ensure numbers are used where expected
  • Reject invalid values – Prevent negative or zero values

If a client provides invalid input, the API should return a clear error message.

Output Validation​

We also validate the model’s output to prevent unexpected behavior.

  • Check for logical errors – Example: No negative arrival times
  • Format responses properly – Ensure consistency
  • Provide error messages – Help clients understand failures

This ensures that clients always receive valid and reliable responses.

Request and Response Models​

To enforce both the input and output validations, we need to define request and response models.

  • Request model – Specifies expected input fields, types, and constraints
  • Response model – Ensures structured, consistent output

These models help standardize communication between clients and the API.

Security and Rate Limiting​

To protect our API, we restrict access and limit requests. The following measures ensure stability and prevent overuse.

  • Authentication – Only allow authorized users
  • Rate limiting/Throttling – Control requests made to prevent abuse

FastAPI is a great tool for building APIs efficiently which provides built-in validation, security, and documentation features.