Machine Learning Engineering: From Basics to Production MLOps

Welcome to the Machine Learning Engineering repository—a structured, comprehensive collection of resources, source code, interactive notebooks, and production deployment patterns designed to teach machine learning from first principles to scalable real-world architectures.

Whether you are beginning your data science journey or mastering production engineering (MLOps), this repository guides you through algorithmic foundations, deep learning, containerization, serverless functions, and Kubernetes orchestration.

⭐ Support this project: If you find these guides, notebooks, and production code helpful, please give this repository a Star on GitHub!

👋 Welcome DataTalks.Club & ML Zoomcamp learners!
Feel free to fork or bookmark this repository as a hands-on companion reference for your homework, midterm projects, and cloud deployments.

Overview

Machine learning transforms data into predictive intelligence and automated decision systems. This repository functions both as an academic syllabus and a practical engineering playbook:

Mathematical Foundations & EDA: Understand linear models, cost functions, loss minimization, and data preprocessing.
Supervised Learning: Master classification and regression with classical algorithms and ensemble tree methods (XGBoost, Random Forests).
Validation & Metrics: Rigorously evaluate models with precision-recall curves, ROC-AUC, confusion matrices, and cross-validation.
Deep Learning & Computer Vision: Build convolutional neural networks (CNNs) and apply transfer learning with TensorFlow/Keras and PyTorch.
Cloud-Native MLOps: Containerize models with Docker, serve REST inference APIs with Flask/Gunicorn, deploy serverless functions, and scale with Kubernetes and KServe.

Video Lectures & Walkthroughs

Accelerate your learning with video lectures and technical walkthroughs corresponding to the curriculum:

📺 Discovering Machine Learning: A Primer Guide	🎓 Neural Network Model Analysis (Harvard CS50 AI)	📺 AI Agents for Health Risk Assessment (Kaggle)

Watch Live Stream	Watch Deep Dive	Watch Walkthrough
Topics: ML definitions, supervised vs. unsupervised paradigms, problem formulation, exploratory data analysis (EDA), data cleaning, feature engineering, and lifecycle.	🎓 Created for Harvard CS50 AI: In-depth breakdown of CNN layer architectures, activation functions, convolutional kernels, pooling, and image classification convergence.	Topics: Clinical interview AI agent on Google Cloud Run, MCP tools integration, real-time risk score inference, and personalized health recommendations.

Featured on Kaggle

🏆 Kaggle VibeCoding Agents Capstone Project:
The work developed in this repository's Heart Disease Risk Assessment project has been featured on Kaggle!

📄 Read the Writeup: AI Agents for Health Risk Assessment on Kaggle

📺 Watch the Video Demo: AI Agents for Health Risk Assessment Walkthrough

Learn how to take a trained tabular machine learning classifier and integrate it into an agentic workflow using the Model Context Protocol (MCP), Gemini (Vertex AI), and Google Cloud Run for conversational healthcare risk assessment.

Learning Roadmap

Follow this progression path to transition from core fundamentals to scalable cloud architectures:

flowchart TD
    subgraph Phase1["Phase 1: Foundations & Classical ML"]
        M01["01: Introduction to ML"] --> M02["02: Regression & EDA"]
        M02 --> M03["03: Classification & LogReg"]
        M03 --> M04["04: Evaluation & Validation"]
    end

    subgraph Phase2["Phase 2: Tree Ensembles & API Deployment"]
        M04 --> M06["06: Decision Trees & XGBoost"]
        M06 --> M05["05: Model Deployment (Flask + Docker)"]
        M05 --> M07["07: Midterm Project Implementation"]
    end

    subgraph Phase3["Phase 3: Deep Learning & Complex Modalities"]
        M07 --> M08["08: Deep Learning & Computer Vision (CNNs)"]
        M08 --> P01["Drug-Drug Interaction (SSP + MLP)"]
        M08 --> P02["Voice Cloning with PyTorch"]
    end

    subgraph Phase4["Phase 4: Cloud-Native & Scalable MLOps"]
        M08 --> M09["09: Serverless Model Hosting (TF-Lite / Lambda)"]
        M09 --> M10["10: Kubernetes Orchestration & Autoscaling"]
        M10 --> M11["11: Production Model Serving with KServe"]
    end

Curriculum & Module Index

Each numbered module contains in-depth documentation, runnable Jupyter notebooks, and production code:

Module	Title & Documentation	Core Topics Covered	Practical Code & Notebooks
01	01-Introduction	What is ML, Core concepts, Supervised learning principles, $Data + Outcome \to Model$, Generalization	homework.ipynb, comparison.ipynb
02	02-Regression	Linear regression, normal equation, exploratory data analysis (EDA), feature engineering, validation split, RMSE	homework.ipynb
03	03-Classification	Binary/multi-class classification, Logistic Regression, Sigmoid function, mutual information, one-hot encoding	homework.ipynb
04	04-Evaluation	Confusion matrix, Accuracy, Precision, Recall, F1 score, ROC and PR curves, AUC, K-Fold cross-validation	homework.ipynb
05	05-Deployment	Serializing models (Pickle/Joblib), building web services with Flask/Gunicorn, Pipenv dependencies, Docker containers	app.py, Dockerfile, homework.ipynb
06	06-Trees & Ensembles	Decision Trees, stopping criteria, Random Forests, Gradient Boosting (XGBoost), hyperparameter tuning	homework.ipynb
07	07-Midterm Project	End-to-end capstone methodology: Problem statement, data preparation, model selection, tuning, and API serving	README.md
08	08-Deep Learning	Artificial neural networks, Convolutional Neural Networks (CNNs), TensorFlow/Keras, image augmentation, transfer learning (includes lecture for Harvard CS50 AI)	homework.ipynb
09	09-Serverless Hosting	Serverless inference on AWS Lambda / Azure Functions, converting models to TensorFlow-Lite, Docker runtime optimization	main.py, Dockerfile, homework.ipynb
10	10-Kubernetes	Container orchestration with K8s, Deployments, Services, Horizontal Pod Autoscaler (HPA), Kind/Minikube local clusters	deployment.yaml, hpa.yaml, q4_predict.py
11	11-KServe	Cloud-native model serving with KServe, automated GPU/CPU scaling, multi-framework support (TF, PyTorch, ONNX), vLLM	quick_install.sh, iris_example.yaml

Featured Capstone Projects

Explore end-to-end case studies implementing machine learning architectures for real-world problems:

1. Heart Disease Risk Assessment (Featured on Kaggle & ozkary.com)

Domain: Healthcare & Preventative Medicine
Dataset: US Centers for Disease Control and Prevention (CDC) EHR data
Architecture: Implements a Model Factory pattern comparing Logistic Regression and Random Forest classifiers, complete with data preprocessing pipelines and containerized REST endpoints.
Publications & Capstone Extension:
Key Assets: README.md, data_train.py, app.py, Dockerfile

2. Drug-to-Drug Interaction (DDI) Classifier

Domain: Pharmacogenomics & Computational Chemistry
Dataset: DrugBank repository of over 190,000 known drug interaction pairs
Architecture: Formulates DDI prediction by extracting SMILES molecular representations, generating Structural Similarity Profile (SSP) vectors, and training Multi-Layer Perceptrons (MLP) for multi-label interaction prediction.
Key Assets: README.md, data_train_mlp.ipynb, app.py, Dockerfile

3. Deep Learning Voice Cloning with PyTorch

Domain: Speech Processing & Generative Audio
Architecture: Audio spectrogram feature extraction, acoustic modeling, neural vocoder pipelines, and PyTorch deep learning implementations.
Key Assets: README.md

4. Additional Exploratory Studies

Vehicle MSRP Price Prediction: Multivariate regression and feature collinearity analysis for car market valuations (regression.ipynb).
End-to-End Census & Demographic Modeling: Socioeconomic classification workflows (end_to_end_machine_learning_project.ipynb).

Model Selection Guide

Selecting the right machine learning algorithm depends on the nature of the target variable, data volume, interpretability requirements, and latency constraints:

Machine Learning Classification

Model Type	Primary Task	Best Suited For	Advantages & Tradeoffs
Linear Regression	Continuous Prediction	Baseline price forecasting, trend lines	Fast, highly interpretable; vulnerable to non-linearities and outliers.
Ridge / Lasso Regression	Continuous Prediction	Datasets with multicollinear or high-dimensional features	Regularization prevents overfitting; balances feature weights.
Logistic Regression	Binary / Multi-class	Disease diagnosis, customer churn probability	Provides calibrated probabilities; linear decision boundary.
Decision Trees	Classification & Regression	Rules-based workflows, demographic segmentation	Natural handling of non-linearities and categorical data; prone to overfitting.
Random Forests	Classification & Regression	Defect detection, credit risk, complex tabular data	Ensemble bagging reduces variance; robust against noise.
Gradient Boosting (XGBoost)	Classification & Regression	Kaggle benchmarks, ranking, fraud detection	State-of-the-art tabular accuracy; requires hyperparameter tuning.
Support Vector Machines (SVM)	Classification & Regression	Quality control, text classification	Effective in high-dimensional spaces with custom kernels.
K-Nearest Neighbors (KNN)	Classification & Regression	Equipment maintenance scheduling, recommendation systems	Instance-based / lazy learning; inference slows with large datasets.
Convolutional Neural Networks (CNN)	Computer Vision	Image recognition, defect inspection, medical scans	Captures spatial hierarchies; requires GPU acceleration and large datasets.
Multi-Layer Perceptrons (MLP)	Complex Multi-modal	Chemical interaction (DDI), non-linear feature maps	Flexible neural modeling; requires careful regularization.

Getting Started

Prerequisites

Python: 3.9 or higher
Package / Environment Manager: pipenv or venv / conda
Docker: For testing containerized models locally

Quickstart Setup

Clone the repository:

git clone https://github.com/ozkary/ml-engineering.git
cd ml-engineering

Create a virtual environment and install dependencies:

# Using Python venv
python3 -m venv .venv
source .venv/bin/activate
pip install jupyter scikit-learn pandas numpy matplotlib seaborn

Or use pipenv inside any module folder (e.g., cd 05-deployment && pipenv install).

Launch Jupyter Lab / Notebooks:
```
jupyter lab
```
Detailed Environment & Setup Guide: For step-by-step instructions on setting up Docker, Pipenv, Kubernetes clusters (Kind/Minikube), and testing APIs, read the complete Getting Started Guide.

Community & Updates

Stay up-to-date with tutorials, guides, and open-source releases:

⭐ Star the Repo: If you find this curriculum helpful, give it a star on GitHub to help other engineers and students discover it!
📬 Newsletter: Sign up for the ML Engineering & AI Newsletter
🐦 Twitter / X: Follow @ozkary
🤖 Companion AI Agents Repo: ozkary/ai-engineering
📊 Companion Data Engineering Repo: ozkary/data-engineering-mta-turnstile

Contributing

Contributions, issues, and feature suggestions are welcome! Whether it is fixing a typo, proposing a new algorithm notebook, or optimizing a Docker configuration, please review our Contribution Guidelines before submitting a pull request.

License

This repository is distributed under the terms of the Apache License 2.0. Feel free to use the code, notebooks, and reference materials in your projects and studies.

ozkary/machine-learning-engineering

Progression mesurée des étoiles

À propos

README.md