Neural Networks

Built an LSTM (Long Short-Term Memory) neural network to predict electric vehicle charger demand across New York City. The project involved processing large-scale spatial-temporal data, implementing one-hot encoding for categorical variables, and establishing a performance baseline using Facebook Prophet. I documented the end-to-end development—from data engineering to model architecture—to share my learning journey and insights. GitHub • Full write-up on Medium

Developed TinyNet, a lightweight deep learning framework built entirely from scratch in Python and NumPy. This project bypasses high-level libraries like PyTorch or TensorFlow to implement core components—including backpropagation, custom activation functions (ReLU, Sigmoid), and various optimization layers—manually. It serves as a transparent laboratory for understanding the calculus and linear algebra that power modern artificial intelligence. GitHub • Full write-up on Medium