Projects.

Independent study into novel ML architectures, measured by computational efficiency and knowledge representation.

Worked with Prof. Amy Hoover on experiments exploring novel machine learning architectures. Implemented and replicated the results of “Topology of Deep Neural Networks” and “Are All Layers Created Equal?”, then tested parameter efficiency in new model types by randomizing or removing layers. Produced visualizations of model decision planes, dataset topology, and model computation/parameter cost.

• PyTorch
• Topology
• Visualization
• NJIT

A Python engine for a hidden-information game, built to train reinforcement learning agents against it.

Reverse-engineered the original game from decompiled source into a complete, efficient Python library — published on PyPI as buckshot-roulette — specifically so AI engines could be developed against it. Used reinforcement learning to search for optimal strategies, monitored training with Tensorboard, and experimented across a range of training algorithms and model architectures.

• Python
• Reinforcement Learning
• Tensorboard
• PyPI

A worldbuilding and mapping tool for sci-fi galaxies — procedural generation with an interactive editor.

Three layers: a Python package that procedurally generates and renders galaxies, a FastAPI backend exposing it, and a Next.js + Pixi.js frontend for interactive visualization. Users generate galaxies with customizable star distributions, then edit the graphs representing transit routes across the world, including resource placement with location and density parameters.

• TypeScript
• Next.js
• Pixi.js
• FastAPI
• Python

Data ingestion and curriculum analysis for a student-built replacement for NJIT's schedule builder.

Contributed to a student project aimed at building a better schedule builder for NJIT. Implemented the data ingestion pipelines and ran a variety of analyses on NJIT course curricula and structure — which meant reverse-engineering Ellucian's Scribe language to get at the degree requirements in the first place.

• Python
• Data Pipelines
• Reverse Engineering

A retrieval-augmented chatbot answering questions about machine learning course material.

Built an LLM-based QA/RAG application to answer questions specific to class material and related topics, as part of my work as a grader. Mined data from approved sources and organized it so answers carry meaningful references — citing Wikipedia or the class website depending on what is most relevant. Used Markprompt for data processing and model access; deployed as a Discord bot in the official class server.

• LLM
• RAG
• Markprompt
• Discord

A VR educational tool letting forensic science students investigate digitized crime scenes.

Worked on a VR tool that lets forensic science students explore and investigate digitized recreations of staged crime scenes. Developed in the NJIT MIXR Lab, first as a class project and then as a summer research project under the NJIT Provost Undergraduate Summer Research Fellowship.

• VR/XR
• C#
• Unity
• NJIT MIXR Lab