Projects

Local-dimming / vision pipeline with a U-Net in PyTorch; C++/CMake core, Dockerized; uses OpenCV and Eigen.

Discord bot that records & transcribes D&D sessions using Azure Speech-to-Text, storing artifacts in Blob Storage.

Developed a diffusion-based model to generate realistic Modbus industrial control system packets for security and intrusion-detection research. Built an end-to-end pipeline to synthesize, serialize, and validate protocol-correct traffic. Designed a statistical validation framework using PCA and SVM classifiers to measure how distinguishable synthetic packets were from real traffic.

An automated cloud based benchmarking framework to evaluate PostgreSQL and ScyllaDB using YCSB. Executed insert, load, stress, soak, and spike workloads on datasets up to 10 million records, collecting throughput, tail latency, and system-level CPU, memory, disk, and network metrics. Developed with modularity in mind, so additional databases can be added in the future with relative ease.

A perceptron-based branch-predictor built on top of Scarab, a cycle accurate CPU simulator. My work is private due to the nature of this assignment, but I've attached a link to Scarab for reference

Scrapes PDFs with PyPDF2 and cleans the text via regex for usable study notes.

A distributed, fault-tolerant, and sharded key-value store built with Flask. It implements causal consistency using vector clocks and manages node membership and replication for high availability.

A reinforcement learning agent built with PyTorch to control a vehicle in the CARLA simulator. The agent uses a policy network to learn how to navigate to a destination while adhering to speed limits.

Modern portfolio built with Next.js + React/TypeScript; Tailwind styling and Dockerized deployment.

A file compression utility using gzip and planned GUI with PyQt5. It features a hash map for categorization and intends to add encryption and automated file categorization using computer vision in the future.

A Python utility to convert HEIC images to JPG format. It uses a pre-trained PyTorch model and the face_recognition library to automatically classify and sort the converted images into categories like 'person', 'animal', or 'landscape'.