ML API: End-to-End Deployment with FastAPI, Docker, GitHub Actions, and AWS EC2

This project showcases a complete machine learning pipeline for predicting healthcare claim attributes, deployed via a production-grade API using FastAPI, Docker, CI/CD, and AWS EC2.

Project Overview

Goal: Predict whether a healthcare claim has a physical copy (has_physical_copy) based on patient and claim metadata.

Tech Stack:

Python, Pandas, Scikit-learn FastAPI + Uvicorn Docker + Docker Hub GitHub Actions for CI/CD AWS EC2 for cloud deployment

Project Architecture

Data Generation → Cleaning → Feature Engineering → Model Training
          ↓
     FastAPI for Prediction Endpoint
          ↓
   Dockerized & CI/CD via GitHub Actions
          ↓
   Deployed to EC2 & Served via Public IP

Features

Synthetic healthcare claim dataset (2M+ simulated records) Robust ETL: missing value handling, anonymization, validation Logistic Regression model with one-hot encoding FastAPI server with /predict endpoint UI for easy testing Dockerized application GitHub Actions CI/CD pipeline Fully deployed to public EC2 server

Folder Structure

.
├── scripts/                  # Data generation
├── etl/                     # Cleaning & validation
├── models/                  # Feature engineering + model training
├── api/                     # FastAPI app + schema
├── notebooks/               # Data exploration
├── Dockerfile               # Image definition
├── requirements.txt         # Python dependencies
└── .github/workflows/       # CI/CD pipeline

How to Run Locally

# Step 1: Generate and clean data
python scripts/generate_data.py
python etl/clean_data.py
python etl/validate_data.py
python models/feature_engineering.py


## Step 2: Train and save model
python models/train_model.py

# Step 3: Run FastAPI server
uvicorn api.main:app --reload
# Visit http://localhost:8000/docs

Dockerization

# Build and run locally
docker build -t project-ml-api .
docker run -p 8000:8000 project-ml-api

GitHub Actions (CI/CD)

Build Docker image
Ready to extend with deployment stages

Dockerhub publishing

# dockerhub publishing
docker tag project-ml-api username/project-ml-api
docker login
docker push username/project-ml-api

AWS EC2 Deployment

Launch EC2 Instance:
AMI: Ubuntu 24.04
Opened ports: 22, 8000, 80, 443
Key Pair: name1.pem

#SSH from Git Bash:
ssh -i name1.pem ubuntu@<your-ec2-public-ip>

# On EC2 (after SSH with p1.pem)
sudo apt update
sudo apt install -y docker.io
sudo usermod -aG docker ubuntu
newgrp docker
docker --version

docker pull username/project-ml-api
docker run -d -p 8000:8000 username/project-ml-api

#Access your API:
Visit http://<ec2-ip>:8000/docs
#here <ec2-ip> is your EC2 instance public ipv4

Instance terminated after demo to avoid AWS charges.

Shutdown (No Charges)

Terminate EC2 from AWS Console:
EC2 Dashboard → Select Instance → Instance State → Terminate
(Optional): Go to EC2 → Volumes → Delete if still attached

Author

Swagath Babu