Archive.

Built GAN-based synthetic medical imagery to learn why generative models fail and how to keep them honest — before the LLM boom made generative AI the default. 2nd place at Brockport's annual AI competition.

• ·Implemented DC-GAN for stable convolutional training and W-GAN to mitigate mode collapse — learned where generative models break and how to keep them honest.
• ·Benchmarked synthetic image quality against real OCT datasets from Kaggle to evaluate against real data, not a demo.
• ·Stress-tested LLMs on high-resolution image processing to find edge cases and hallucination triggers — the same failures I now watch for in production agent systems.

• Python
• PyTorch
• DC-GAN
• W-GAN
• Kaggle

Built OCR pipelines to process high-resolution image datasets — classical computer vision paired with Gemini API for context-aware recognition, learning where models break on edge cases.

• ·Built pipelines to handle diverse, complex visual data — the kind of edge-case-heavy work that teaches you where models break.
• ·Used OpenCV, NumPy, and Matplotlib for the image processing layer — the fundamentals that matter when the fancy API fails.

• Python
• Gemini API
• OpenCV
• NumPy
• Matplotlib

Built a deep learning pipeline to detect scientific image forgery — instance segmentation on real competition data. 2nd place at SUNY Brockport's annual AI competition.

• ·Implemented Mask R-CNN architecture in PyTorch and Torchvision for object detection and instance segmentation.
• ·2nd place at SUNY Brockport's annual AI competition.

• Python
• PyTorch
• Torchvision
• Mask R-CNN

Built an AI-powered study companion service for SUNY Brockport to ensure ADA compliance for digital content — led an 8-person team through real Agile sprints to ship a system for real institutional deployment.

• ·Led an 8-person team through Agile sprint cycles to ensure efficient delivery and address complex implementation challenges in a Next.js/TypeScript environment.
• ·Architected the backend to handle real institutional requirements — Clerk auth, Prisma ORM on Supabase, file handling with Supabase Buckets, and an AI layer using OpenAI API + pgVector for retrieval.
• ·Built an accessible ALT text generator for faculty uploading PowerPoints to ensure ADA Title II compliance (mandated April 2026) — still in active use today, over 100 users and consistently positive feedback.
• ·Presented at CCSCNE 2025 and Scholars Day.

• Next.js
• TypeScript
• Clerk
• Prisma
• Supabase
• OpenAI
• pgVector