This repository contains a curated collection of applied machine learning and deep learning projects implemented from scratch and with standard frameworks.
The goal of this portfolio is to explore core ML/DL domains through practical, well-structured, and reproducible projects.
The projects emphasize:
- solid problem formulation
- clean data pipelines
- model design and training
- rigorous evaluation
- clear documentation and analysis
Each subdirectory focuses on a specific domain of machine learning:
├── classical_ml
├── computer_vision
├── nlp
├── README.md
└── recommender_systems
Each project follows a consistent structure:
- data loading and preprocessing
- model implementation
- training and evaluation
- discussion of results and limitations
To set up the environment and install dependencies, run:
uv sync
Supervised learning on tabular data with an emphasis on:
- feature engineering
- model comparison
- cross-validation
- interpretability
Typical models: Linear/Logistic Regression, Random Forests, Gradient Boosting.
Image-based tasks using convolutional neural networks.
Topics include:
- CNN architectures
- overfitting and regularization
- data augmentation
- performance analysis
Framework: PyTorch.
Text classification and representation learning using classical NLP techniques and neural models.
Topics include:
- Bag-of-Words and TF-IDF
- text embeddings
- sequence models
Personalization and ranking problems.
Approaches include:
- content-based filtering
- collaborative filtering
- similarity-based ranking
Unsupervised and semi-supervised detection of abnormal patterns.
Techniques include:
- Isolation Forest
- One-Class SVM
- Autoencoders
- Python
- NumPy, pandas, scikit-learn
- PyTorch
- matplotlib / seaborn
- Jupyter Notebook
This repository is intended to:
- build strong practical intuition in ML/DL
- compare different domains to identify specialization interests
- serve as a technical portfolio for internships or junior roles
Each project is self-contained and documented with clear assumptions, design choices, and results.
This is an evolving repository. New projects and improvements are added progressively as skills deepen and more advanced topics are explored.