An end-to-end machine learning pipeline comparing statistical, tree-based, and deep learning architectures to predict ridership for Mexico City's Cablebús system.

• The Core Tech: LightGBM, PyTorch (LSTM), Facebook Prophet, Statsmodels (SARIMA), Optuna.
• Architectural Strategy: Implemented isolated, line-specific models to capture unique demographic flows. Developed a direct multi-model forecasting horizon approach (separate models per lead day) to explicitly prevent the compounding error propagation typical of recursive time-series setups.
• Impact: Provides stable uncertainty quantification for long-term urban risk management and operational planning.

statistical analysis and deep learning applied to financial market indicators and forecasting.

• The Core Tech: Python, Pytorch, Scikit-Learn, Pandas, Advanced Feature Engineering.
• Architectural Strategy: Built a robust validation pipeline specifically tailored for non-stationary financial data, incorporating strict walk-forward cross-validation to eliminate data leakage and optimize feature importance metrics under high-noise conditions.

An engineering exploration into leveraging massive, unlabeled video datasets to train deep learning architectures for pixel-level motion estimation.

• The Core Tech: PyTorch/OpenCV, FlowNet Simple, YouTube-8M Dataset, Motion Analysis Fundamentals.
• Architectural Strategy: Developed a custom modification over the FlowNet Simple backbone to accelerate convergence and optimize inference metrics. Built a pipeline targeting the complexities of training on raw web video, addressing challenges in frame-rate sampling dynamics (large vs. small displacements) and identifying data-cleansing strategies to isolate noise like scene cuts and text overlays.

An educational and rigorous implementation of neural network fundamentals written purely in Java without external ML libraries.

• The Core Tech: Java (Object-Oriented Architecture), Matrix Mathematics.
• Architectural Strategy: Built backpropagation, weight initialization schemes, activation functions, and gradient descent optimization completely from first principles. This repository highlights a deep, foundational understanding of memory allocation, matrix operations, and the core mathematical mechanics behind modern AI frameworks.

• Languages: Python, Java, Bash, SQL, LaTeX.
• Frameworks & AI: PyTorch, LightGBM, Optuna, Scikit-Learn, Prophet, OpenCV.
• MLOps & Infrastructure: Poetry, Git, Docker, CI/CD Pipelines.
• Specializations: Time-Series Forecasting, Deep Generative Learning, Distributed Systems Concepts.

• LinkedIn: linkedin.com/in/diegopaez * GitHub: @DiegoEPaez