Projects

Orchestrated a multi-agent workflow in Python on RHEL Linux using academy-agents, PBS Pro, Intel MPI, LangChain, FastMCP, and Parsl, running parallel, distributed agents across nodes of Aurora, an exascale supercomputer.

The workflow abstracts away the technical details, so non-technical researchers can run HPC jobs themselves.

Implemented the STL containers vector, string, and their iterators against the standard library specification.

Wrote CUDA kernels to perform triangle stripping on 3D models.

Parallelized sections of a C++ image processing tool using CUDA.

CUDA: wrote kernels for blurring, dilation, erosion, and binarization, using shared memory to cache image regions that had already been read.

Benchmarking: compared the CUDA kernel times against the original CPU times, then used NVIDIA Nsight Compute to find the most efficient kernel configuration for my GPU architecture.

Built an AI agent workflow that writes unit and integration tests for an existing codebase, using MCP tools and skill files.

Built a tool that parses the OWL2 ontology format and converts it into the custom ontology format used by our compiler’s knowledge base.

Machine learning work mapping the medullary reticular formation in the developing mouse brain, to help uncover the neural control of orofacial behaviors. This research became my first-author ICBCB 2026 paper.

Worked with a team to build a full-stack application that predicts gentrification in Chicago. I scraped permit data, median income, crime rates, and other features from the Chicago Data Portal and other city government sites, then cleaned and normalized the data and fed it into a linear regression model.

Frontend: integrated the Google Maps UI into the site, with custom markers.

Backend: in Django, set up the REST API, the core logic, the database design, and containerization with Docker Compose.

Built an order matching system in Java using common object-oriented design patterns, with error handling throughout.

Built a website that displays live foreign exchange conversion rates.

Extracted and cleaned thousands of transactions from Arifpay’s POS tool using MySQL and Pandas, then applied time-series methods (an ARIMA model) to the historical data to forecast merchant profit and inform Arifpay’s pricing decisions.

Used alpha-beta pruning to find the best move in a game of tic-tac-toe.

Built an app with a friend to help users track gym workouts, as alpha testers for ChatGPT plugins. The plugin feature was later scrapped by OpenAI.

Worked in a team of 6 to build a computer vision tool that analyzes facial expressions, aiding people with social-emotional agnosia during real-time conversation.

Frontend: built the homepage UI in Next.js, and modeled a 3D asset displayed through Three.js (eventually cut).

An Alice in Wonderland themed maze game.

Game development: wrote the core logic for coin collection and character collision.

3D modeling: modeled every character and the full environment from scratch.

Full-stack work alongside a senior software engineer.

Diagrams: designed the sequence and entity-relationship diagrams defining how attorneys, the calendar, and users relate to each other.

Backend: built the core business logic in Spring Boot, connected to MySQL through Hibernate ORM.

Frontend: built the UI in Next.js and wrote the REST endpoints so users could submit consultation requests.

Testing: exercised the endpoints in Postman to confirm they reject malformed input.

Built a high-resolution version of the guitar, handled the booleans and mesh cleanup in ZBrush, then retopologized it in Blender.