Machine Learning: From First Principles to Production

A curated collection of projects covering the complete Machine Learning lifecycle—from core algorithms and data engineering to experiment tracking, MLOps, and deploying production-ready machine learning systems.

ML Fundamentals Learn Machine Learning from first principles by implementing data preprocessing, exploratory data analysis, feature engineering, supervised learning algorithms, model evaluation, hyperparameter tuning, pipelines, and complete end-to-end machine learning workflows.
Data Engineering Build data engineering workflows on AWS by creating ETL pipelines with S3, Glue, Lambda, and related cloud services to automate data ingestion, transformation, and preparation for production machine learning systems.
Git for ML Master Git for machine learning projects by learning branching strategies, commits, merge conflicts, pull requests, version history, collaboration workflows, and production-ready source code management.
Data Versioning Learn how to version datasets, models, and machine learning pipelines using DVC while integrating with Git to create reproducible experiments, automate workflows, and manage large data efficiently.
MLflow Track machine learning experiments using MLflow by logging parameters, metrics, artifacts, and models, managing experiment runs, registering models, and integrating with DagsHub and Databricks.
Docker Containerize machine learning applications using Docker by building images, managing containers, packaging project dependencies, and deploying reproducible ML services across different environments.
Continuous Integration Automate machine learning development using GitHub Actions by configuring Continuous Integration workflows that perform testing, validation, linting, and quality checks whenever code changes are pushed.
End-to-End MLOps Build a modular MLOps pipeline from scratch by implementing data ingestion, feature engineering, configurable DVC pipelines, experiment management, logging, parameterization, and cloud-based data versioning.
Production Project In this project, I first set up the project structure, created a virtual environment, installed all the required dependencies, and configured local packages. Then I created a MongoDB Atlas database, uploaded the dataset from a Jupyter notebook, and connected my project to fetch the data. After that, I implemented logging and exception handling to make debugging easier. The pipeline then performs data ingestion by loading data from MongoDB, data validation to check the dataset quality and schema, data transformation to preprocess the data for machine learning, and model training to build the prediction model. Finally, I configured AWS S3 as a model registry, where the newly trained model is evaluated against the existing production model using the F1-score. If the new model performs better than the existing one, it is automatically uploaded to the S3 bucket; otherwise, the current production model is retained. This creates a complete end-to-end MLOps pipeline from data storage to automated model deployment.