A PyTorch-based neural network with a full end-to-end ML pipeline to optimize pit stop strategy, leveraging Formula 1 telemetry data and accurately forecasting remaining tire life based on telemetry data. Used large scale data sourcing (using FastF1), data cleaning, data ingestion, preprocessing, custom feature engineering, circuit clustering (K-means), and hyperparameter optimization (using Optuna).

A custom PyTorch-based neural network for image classification, featuring parallel convolutional branches, advanced ML techniques (MixUp, CutOut, LR scheduling), and a preprocessing/augmentation pipeline with expert-guided attention mechanisms and achieving 90%+ accuracy.

Web-based Workout Tracking. I didn't like any of the existing apps for workout tracking. So developed a custom workout tracking web app with a mobile-friendly interface using Next.js, React, TypeScript, and Tailwind CSS.

A suite of Python-based CLI tools for quantitative financial analysis, including backtesting elementary trading strategies, risk assessments, and technical analysis.

Developed a web-based synthesizer application simulating a classic subtractive synthesizer. Has two oscillators, a modifiable filter, an LFO, and even multiple effects (reverb, delay, chorus). Personal favourite project.