Projects
A gallery of systems, applications, and tools I have engineered.
ADAPT - Automated Parallel Training for ML Workloads on Heterogeneous Systems
An implementation of an efficient, adaptive, and optimized parallel training strategy. Utilizes Data, Model, and Pipeline Parallel training formats, with experimental support for Tensor Parallel training. Automates training script layouts by evaluating systems-level cost functions on heterogeneous clusters.
Classification of Malicious URLs using Ensemble Machine Learning
Built a cascading classifier utilizing 7 machine learning models to detect malicious URLs. Six base models evaluate independent feature sets (domain attributes, symbol frequencies, DOM structure, and text content), passing predictions to a final decision tree ensemble. Published at the IEEE-CSDE'21 conference.
Movie Success Prediction using Machine Learning
An exploratory study comparing statistical approaches to predict box-office success. Conducted extensive data cleaning and analysis to establish correlation factors (actors, budget, rating, director), implementing multiple models on data subsets. Published at the MLDS'21 conference and in the IJAIA'21 journal.
Classification of Habitable Exoplanets
Developed model layouts to identify habitable exoplanets based on NASA's Kepler dataset. Extracted planetary and host star properties (mass, radiation, orbit, composition) and evaluated model recall profiles to establish optimal predictors for astrobiological habitability.