Projects

AI agents, Vanderbilt University

Three agents built in fall 2026 for Software Engineering in the Age of AI (CS 8395) at Vanderbilt, taught by Professor Jules White. Each one calls language models through Vanderbilt's Amplify platform, and each follows the same idea: the model makes the judgment calls, and ordinary code checks its work before anything reaches the person using it.

GrantScout

Finds research grants that suit a university lab. The model screens and ranks opportunities in parallel; every reply is checked against a schema, and the model is asked to correct any that fail. Repeated calls are cached, failed calls are retried, and a hard spending limit keeps a run within budget.

Python

Iconoclast

Rearranges the icons on a Windows desktop so they never cover the subject of the wallpaper. A language model judges how important each file is; a vision model, with offline face and sun detection as a check, finds the subject of the image; and a layout solver moves the icons around it. It works through the Windows system interfaces directly, saves an undo snapshot before changing anything, and confirms that each icon landed where intended.

Python · Windows · 46 automated tests

Freeloader

Reads a student's course syllabi and finds the free student edition of each paid tool the courses require. The model must quote the syllabus for every tool it lists, and any quote that cannot be found in the document is discarded. Offers come only from a researched catalog of sixty, so nothing is invented, and a small local dashboard adds up the savings as files are read. Built and tested; not yet run against the live service.

Node.js with a Python worker · 34 automated tests

I am happy to walk through any of them in more detail.