Projects

A clinician pastes the discharge note they already wrote, and Aperta turns it into what the patient still understands at home: a hand-drawn body with a red spot on the place, one plain sentence for what happened, every medicine with what it is for and how to take it, what to do next, and step-by-step walkthroughs for the things a note assumes you already know, like rinsing a tooth socket with a syringe or using an inhaler with a spacer. In the patient's language, read aloud one idea per screen, and printed on one sheet for the fridge. Then the clinician turns the screen around. The note parser, the library of 205 conditions in 21 categories, drug resolution and duplicate-ingredient detection through RxNorm, and a prescriber check that quotes the FDA label are deterministic code running in the page itself, so the published site needs no server and no API keys. Nothing clinical is generated: a model only translates or fills a parsing gap, and any line that cannot be read as a drug order is handed back to the clinician under "Not used" rather than guessed at. All eighteen anatomical views and nine walkthroughs are drawn by hand. Built at VTHacks 14 with Lauren Kim.

A whole-brain Drosophila simulation built to test whether a learned memory reactivates on its own during sleep. 144,209 neurons and 21,868,062 connections from the male CNS connectome, integrated cell by cell in Brian2, with an odour memory encoded by dopamine-gated plasticity at Kenyon cell to output neuron synapses, followed by an offline period with no sensory input and a search for the same ensemble switching itself back on, scored against a degree-preserving shuffled connectome, an unlearned-weights control and size-matched random ensembles. The viewer renders all 126,109 soma positions in 3D and lights the neurons that spike, and every figure is read from the simulation's own output files at page load. Negative results are reported in full, including two published cells the pipeline found to be mis-specified and a positive control establishing that the published model cannot represent a discrete episode at all.

Beautiful, full-featured weather app designed for fast, accurate daily decisions, with current conditions, hourly trends, multi-day forecasts, air quality, wind, humidity, pressure, UV, visibility, sunrise and sunset details, and severe-weather awareness in one polished interface.

A privacy-first system monitor for macOS with companion iPhone and Apple Watch apps. Live menu bar stats, real-time CPU, GPU, and memory charts, battery health trends, and thermal insights with fan control on Apple Silicon, plus an MCP server and CLI that let coding agents check system capacity before running heavy tasks. No cloud, no telemetry; everything stays on device.

A 3D graphing calculator that runs entirely in the browser. Plots function, cylindrical, spherical, parametric, and implicit surfaces alongside space curves and vector fields, with click-to-inspect calculus tools for tangent planes, symbolic partial derivatives, gradients, divergence and curl, level curves, Riemann sums, and animated Frenet frames. The expression parser, symbolic differentiation, and marching cubes geometry are written from scratch with no dependencies.

A one-command, fully local video-to-3D pipeline for Apple Silicon Macs. Point it at ordinary phone videos of a place and it reconstructs an interactive Gaussian splat scene you can fly around in a browser, running ffmpeg, COLMAP, and Brush end-to-end with no cloud or GPU rig.

A tiny, zero-dependency macOS menu bar app for ejecting external drives, SD cards, SSDs, and mounted disk images without digging through Finder. One-click eject, live volume count badge, type-aware icons, and an Eject All shortcut, all in a single Swift file.

Synthesize distributed intelligence from ambient compute with instantaneous inference across edge devices.

Open-source, browser-based communication tool for people with ALS using real-time eye tracking to select letters and speak them aloud.

Leading Oakton HS’s premier coding club, organizing competitive programming, web dev, and USACO/LeetCode prep while mentoring members.

Developing compact, efficient PCBs for USB-C storage and embedded applications with a focus on signal integrity and thermal performance.

Reverse-engineered Apple Silicon to model CPU/GPU/Neural Engine integration, cache hierarchies, and power efficiency strategies.

Assignments for the Oakton High School AV CS Data Structures & Algorithms course.