API: /predict Endpoint¶

The /predict endpoint accepts penguin measurements and returns a predicted species.

The server must be running locally to use this endpoint. See the project README for instructions on how to start the server.

Endpoint¶

POST http://127.0.0.1:8000/predict

Request¶

Content-Type: application/json

Field	Type	Description
bill_length_mm	float	Bill length in millimeters
bill_depth_mm	float	Bill depth in millimeters
flipper_length_mm	float	Flipper length in millimeters
body_mass_g	float	Body mass in grams

Example Request¶

{
  "bill_length_mm": 39.1,
  "bill_depth_mm": 18.7,
  "flipper_length_mm": 181,
  "body_mass_g": 3750
}

Response¶

Field	Type	Description
prediction	string	Predicted species label

Example Response¶

{ "prediction": "Adelie" }

Error Response¶

If a required field is missing or non-numeric, the server returns HTTP 400:

{ "detail": "Missing required feature: 'bill_depth_mm'" }

Interactive Docs¶

When the server is running, FastAPI provides interactive documentation at:

http://127.0.0.1:8000/docs

Free Server Hosting Options¶

Many hosting options require a credit card to sign up, even if you can host your function for free.

Possible options for hosting a free live endpoint:

Render - simplest deploy, free tier, point at a GitHub repo and it builds automatically
Railway - free tier, similar to Render, good developer experience
Fly.io - free tier, more control, slightly more setup
Hugging Face Spaces - free, no credit card, popular in ML circles, supports FastAPI

Cloud Hosting Options¶

The major cloud providers all offer free tiers that can host a FastAPI endpoint, but all require a credit card to sign up (even for free usage).

AWS Lambda + API Gateway - 1M requests/month free; FastAPI works via the Mangum adapter
Azure Functions - 1M requests/month free; FastAPI works via a similar adapter
Google Cloud Run - 2M requests/month free; easiest of the three for FastAPI; build a Docker container and deploy

For this, Render or Hugging Face Spaces are the easiest starting points. For production experience closest to what industry uses, Google Cloud Run may be a good option to explore.