🚀 Model Serving — FastAPI + Docker + Cloud Run

Part of the MLOps Portfolio Series — Project 5 of 10
Deploys the XGBoost champion model from Project 4 as a production FastAPI service — containerised with Docker, tested with pytest, and ready for Cloud Run deployment.

---

📂 Project Resources

Resource	Link
🚀 FastAPI App	src/api/app.py
🏋️ Model Trainer	src/model/train.py
🧪 API Tests	tests/test_api.py
🐳 Dockerfile	Dockerfile
🤖 CI/CD Workflow	.github/workflows/serving.yml
📋 Requirements	requirements.txt

---

🎯 What This Project Does

Takes the champion XGBoost model from Project 4 and puts it in front of real users:

1. Trains champion model on real UCI Bank Marketing data (41,188 rows)
2. Serves predictions via FastAPI REST endpoints
3. Containerises with Docker — multi-stage build, non-root user, health check
4. Tests all endpoints with pytest (11 passing tests)
5. CI/CD pipeline trains → tests → builds Docker → tests container on every push

---

📊 Model

Property	Value
Algorithm	XGBoost (tuned with Optuna — Project 4)
Dataset	UCI Bank Marketing (41,188 rows, real data)
AUC	0.8176
Baseline AUC	0.8174 (Project 1 RandomForest)
Features	19 (duration dropped — leakage)
Target	Term deposit subscription (binary)

---

🔌 API Endpoints

Method	Endpoint	Description
GET	/	Welcome message
GET	/health	Health check + model info
GET	/model-info	Full model metadata
POST	/predict	Single customer prediction
POST	/predict-batch	Batch predictions

Example Request

curl -X POST "http://localhost:8000/predict" \
  -H "Content-Type: application/json" \
  -d '{
    "age": 42,
    "job": "admin.",
    "marital": "married",
    "education": "university.degree",
    "default": "no",
    "housing": "yes",
    "loan": "no",
    "contact": "cellular",
    "month": "may",
    "day_of_week": "mon",
    "campaign": 2,
    "pdays": 999,
    "previous": 0,
    "poutcome": "nonexistent",
    "emp.var.rate": 1.1,
    "cons.price.idx": 93.994,
    "cons.conf.idx": -36.4,
    "euribor3m": 4.857,
    "nr.employed": 5191.0
  }'

Example Response

{
  "prediction": 0,
  "probability": 0.0823,
  "label": "No — will not subscribe",
  "confidence": "High",
  "model_version": "XGBoost-v1"
}

---

🏗️ Architecture

mlops-model-serving/
├── src/
│   ├── api/
│   │   └── app.py          # FastAPI app — 5 endpoints
│   └── model/
│       └── train.py        # Train + save artifacts
├── tests/
│   └── test_api.py         # 11 pytest tests
├── artifacts/              # model, feature_order, encoders, model_info
├── Dockerfile              # Multi-stage build
├── .github/
│   └── workflows/
│       └── serving.yml     # CI: train → test → docker build → test container
├── requirements.txt
└── Makefile

---

🚀 Quick Start

git clone https://github.com/jumma786/mlops-model-serving.git
cd mlops-model-serving
pip install -r requirements.txt

# Train model on real data
python src/model/train.py --data-path data/bank-additional-full.csv

# Run tests
make test

# Start API server
make run
# Open http://localhost:8000/docs

---

🐳 Docker

# Build
make docker-build

# Run
make docker-run
# API available at http://localhost:8000

---

☁️ Cloud Run Deployment

# Build and push to GCR
gcloud builds submit --tag gcr.io/PROJECT_ID/mlops-model-serving

# Deploy to Cloud Run
gcloud run deploy mlops-model-serving \
  --image gcr.io/PROJECT_ID/mlops-model-serving \
  --platform managed \
  --region europe-west2 \
  --allow-unauthenticated

---

🔗 MLOps Portfolio Series

#	Project	Repo	Status
1	Multi-Model Tournament Pipeline	mlops-model-tournament	✅
2	Scheduled Retraining + DVC + MLflow	mlops-retraining-pipeline	✅
3	Feature Engineering as Versioned Artifact	mlops-feature-pipeline	✅
4	Hyperparameter Tuning with Optuna + MLflow	mlops-hyperparameter-tuning	✅
5	FastAPI + Docker + Cloud Run	mlops-model-serving	✅ This repo
6	Feature Store with Feast + Redis	mlops-feature-store	🔜
7	Model Monitoring & Drift Detection	mlops-model-monitoring	🔜
8	A/B Testing Framework	mlops-ab-testing	🔜
9	Airflow Pipeline Orchestration	mlops-airflow-pipeline	🔜
10	Kubernetes ML Platform	mlops-k8s-platform	🔜

---

📝 Key MLOps Concepts Demonstrated

• REST API serving — FastAPI with Pydantic validation
• Containerisation — Docker multi-stage build, non-root user
• CI/CD for serving — train → test → build → test container
• Input validation — Pydantic schemas reject invalid inputs
• Health checks — Docker HEALTHCHECK + /health endpoint
• Batch inference — /predict-batch for high-throughput use cases

---

👤 Author

Jumma Mohammad Teli — Data Analyst & ML Engineer
📍 Birmingham, UK
📧 jummamohammad477@gmail.com
🔗 LinkedIn | GitHub

---

Project 5 of 10 — MLOps Portfolio Series. Builds on Project 4 (Hyperparameter Tuning) by deploying the champion model as a production API.