MLOps Mini Project

Concept:
A lightweight, end-to-end MLOps pipeline demonstrating how to train, serve, containerize, and deploy a machine learning model using FastAPI for inference and Streamlit for visualization, orchestrated through Docker and automated with GitHub Actions CI/CD.

Problem Solved:
Shows how to take a simple ML model from training to production with proper structure: API serving, frontend UI, Dockerized microservices, automated builds, and repeatable workflows.

Scope:

• Input:
  CSV dataset used for training a scikit-learn model.

• Process:
  A minimal but complete MLOps workflow:

  1. Model Training: Train a model using scikit-learn and serialize with joblib.

  2. Backend Inference API (FastAPI): Load model, serve predictions via REST.

  3. Frontend Dashboard (Streamlit): Interactive UI for sending predictions and viewing results.

  4. Containerization: Dockerfiles for backend and frontend, orchestrated using Docker Compose.

  5. CI/CD: GitHub Actions pipeline that builds images and validates code on every push or PR.

• Output:
  REST API docs, interactive dashboard, working Docker microservices, and CI/CD pipeline validation.

Features:

• Simple reproducible MLOps template

• Train → Serve → Visualize pipeline

• FastAPI backend for inference

• Streamlit dashboard for UI interaction

• Docker Compose for local deployment

• GitHub Actions workflow for automated builds

• Organized project structure (API, frontend, models, src)

• Ready for extension into a larger MLOps system

Tech Stack:

• Backend: FastAPI

• Frontend: Streamlit

• ML Framework: Scikit-learn

• Language: Python 3.12

• Containerization: Docker, Docker Compose

• CI/CD: GitHub Actions

• Utilities: pandas, numpy, joblib

Project Structure (High Level):

• src/ — Training script

• models/ — Saved trained model

• app/ — FastAPI backend

• frontend/ — Streamlit dashboard

• .github/workflows/ — CI/CD pipelines

• docker-compose.yml — Service orchestration

• Dockerfile.fa — Backend container

• frontend/Dockerfile.dashboard — Frontend container

Performance:
(Light project — no heavy metrics)

• FastAPI response time: low-latency JSON inference

• Streamlit dashboard loads instantly

• Dockerized startup < 15 seconds

• CI/CD builds containers automatically

Future Improvements:

• Add monitoring (Prometheus + Grafana)

• Deploy using Kubernetes

• Add automated model retraining

• Add unit/integration tests

• Publish images to Docker Hub / AWS ECR