Redesigned the University of Dayton Makerspace's student check-in and safety training verification system from the ground up, replacing a legacy Microsoft Access-based workflow with a modern cross-platform mobile application deployed on an iPad kiosk. The app allows students to sign in by name or UD ID, select a visit reason and work area, and automatically verifies their completed safety certifications by syncing directly with the Canvas LMS API. Students lacking required training for a selected area are blocked from signing in for that space. On sign-out, students confirm workspace cleanup before their session is closed. Upon successful sign-in, the system sends a print job over the local network to a Raspberry Pi running a custom FastAPI service, which generates a PDF badge via ReportLab and sends it to a CUPS printer — the badge displays the student's name and training-credential icons for at-a-glance verification by staff. On the administrative side, the app provides a password-protected dashboard with student record management, bulk CSV/XLSX import, attendance history, database maintenance tools, and a statistics screen with bar and line charts showing visit frequency, average stay duration by day of week, work area utilization, and visit-reason distribution — all exportable to CSV. All data is stored locally on-device using AsyncStorage, with Canvas training completions synced automatically on a configurable interval. The project was built end-to-end using Cursor AI as the primary development environment.
Converted an unpipelined MIPS processor simulator written in C into a 5-stage pipelined simulator with explicit IF, ID, EX, MEM, and WB pipeline registers. Implemented core pipeline behavior including data forwarding, load-use stalls, branch/jump handling with pipeline flushes, and correct write-back/memory behavior, while preserving architectural rules like keeping R0 constant and checking for invalid memory accesses. The simulator assembled and executed a small MIPS-like instruction set including arithmetic, loads/stores, branches, and jumps, and reported execution statistics such as cycle count, IPC, and CPI.
Built a Python deep learning pipeline to detect malaria in microscopy cell images using the NIH malaria dataset. Designed and evaluated a custom VGG-style CNN and compared it against a fine-tuned ResNet-50 transfer learning model, using confusion matrices, ROC curves, and accuracy analysis to measure performance. Developed a preprocessing pipeline with noise reduction, luminance equalization, and normalization, and tested the effects of preprocessing and data augmentation on model generalization. The custom CNN achieved the strongest baseline result at 95.81% accuracy, slightly exceeding the transfer learning model, while additional preprocessing and augmentation proved detrimental in this dataset.