Contributing to the Twilio Flex enterprise contact center platform supporting 20k+ concurrent agents — real-time call handling, compliance workflows, and performance analytics. Building backend logic and TypeScript/React UI for a Fair Credit Reporting Act disclosure system handling 4% of eligible calls at a 10M+ monthly-call center. Refactoring event fanout systems, simplifying architecture and reducing tech debt.
Founder and lead of engineering for a hotel management platform serving independent properties — owning architecture, technical hiring, and product direction. Built the initial product in Java with Spark and Twilio. Selected for Purdue's VentureX accelerator with $5k funding.
Built a VR cave simulation in Unity rendering LiDAR-collected geological data into an immersive walkthrough for classroom instruction and public outreach. Designed interactive elements that let students explore inaccessible cave systems, deployed across multiple Purdue outreach events.
Designed and implemented a Conditional GAN (Python, PyTorch, NumPy) generating synthetic chest X-ray images for medical education. Validated outputs against ground-truth CT data; presented at the Summer Undergraduate Research Symposium.
My favorite project — doesn't use any real data, but works beautifully. Single-threaded matching engine hitting 2.43M orders/sec at sub-microsecond latency (p50 300ns). Price-time priority matching, partial fills, O(1) cancellation via hashmap-iterator locator, and IOC/FOK/Market order types. Exposed to Python via pybind11; validated with a 29-test pytest suite covering edge cases and invariant checks.
Real-time arbitrage detector pulling live order-book data from 5+ exchanges via CCXT, flagging fee-aware profit opportunities. Backtested strategies on simulated paper trades across thousands of trades.
A bootloader and kernel with basic file read/write/edit in a simulated RAM disk. Implemented VGA text-mode output and PS/2 keyboard input via direct hardware ports, plus the real-mode → protected-mode transition with GDT setup.
Trained a Conditional GAN to generate synthetic chest X-rays conditioned on pathology labels (No Findings, Mass, Pneumonia), paired with an SRGAN to upscale outputs to diagnostic quality. Designed specifically for low memory — runs on a GTX 1050Ti, where most GANs can't. Presented at the SURE Symposium at UNE with Dr. Sylvain Jaume.
Interactive medical image annotation tool — place points, connect them into outlines, and export the result as an STL file for 3D modeling. Built for my chest X-ray research at UNE to explore segmentation approaches before moving to GANs.