Inference API Overview

Introduction

The Inference API is a FastAPI-based service that provides credit risk prediction capabilities using a deep learning model. The API predicts whether a customer’s credit risk is “good” or “bad” based on various customer attributes.

Base URL

http://localhost:8000

API Information

title

string

Credit Score Prediction API

version

string

1.0.0

description

string

API to predict the credit risk of a customer using a deep learning model

Authentication

Currently, the API does not require authentication. It is configured with CORS middleware to allow requests from any origin (*).

In production environments, you should limit the allowed origins to specific trusted domains for security.

CORS Configuration

The API is configured with the following CORS settings:

Allow Origins: * (all origins)
Allow Credentials: true
Allow Methods: * (all methods)
Allow Headers: * (all headers)

Architecture

The API uses a singleton pattern for model inference, ensuring efficient resource usage:

Model Loading: The deep learning model is loaded once at startup from model_weights_001.pth
Preprocessor: A scikit-learn preprocessor (preprocessor.joblib) handles feature engineering
Configuration: Model architecture is defined in model_config_001.yaml
Inference Engine: PyTorch-based neural network runs on CPU for predictions

Interactive Documentation

FastAPI automatically generates interactive API documentation:

Swagger UI: Available at /docs
ReDoc: Available at /redoc
Root Endpoint: Redirects to /docs

Client Examples

import requests

base_url = "http://localhost:8000"

# Make a prediction request
payload = {
    "Age": 35,
    "Sex": "male",
    "Job": "unskilled and resident",
    "Housing": "free",
    "Saving accounts": "NA",
    "Checking account": "NA",
    "Credit amount": 9055,
    "Duration": 36,
    "Purpose": "education"
}

response = requests.post(
    f"{base_url}/credit_score_prediction",
    json=payload
)

result = response.json()
print(f"Prediction: {result['prediction']}")
print(f"Probability: {result['probability']}")

Error Handling

The API returns standard HTTP status codes:

200: Successful prediction
422: Validation error (invalid input data)
500: Internal server error during inference

Show Example Error Response

{
  "detail": "An error occurred during inference: <error message>"
}

Running the Server

Start the API server using uvicorn:

uvicorn server.api:app --reload --port 8000

Or with uv:

uv run uvicorn server.api:app --reload --port 8000

Inference API

Model Architecture

Data Processing

Training

Introduction

Base URL

API Information

Authentication

CORS Configuration

Architecture

Interactive Documentation

Client Examples

Error Handling

Running the Server

Next Steps

Credit Score Prediction

API Schema

Build docs developers (and LLMs) love

Inference API

Model Architecture

Data Processing

Training

​Introduction

​Base URL

​API Information

​Authentication

​CORS Configuration

​Architecture

​Interactive Documentation

​Client Examples

​Error Handling

​Running the Server

​Next Steps

Credit Score Prediction

API Schema

Build docs developers (and LLMs) love

Introduction

Base URL

API Information

Authentication

CORS Configuration

Architecture

Interactive Documentation

Client Examples

Error Handling

Running the Server

Next Steps