Designing solutions to complex technological systems.

A free tool that turns a written script into a finished narrated video, with voiceover, drawn visuals, subtitles, and music, running entirely on an ordinary laptop.

Chose code-drawn frames over AI video so it needs no graphics card or per-video fees, cached each stage on content so editing one line redoes only that line, and measured each performance change, cutting a 100-second render from four and a half minutes to two and a half.

Python, ffmpeg, matplotlib, edge-tts

Extended the shrine-tracking method to city blocks: a Chrome extension and local Python server that compare the oldest and newest Street View imagery along a marked block and report whether each building frontage improved, declined, or held steady.

Ran the vision model entirely on the user's machine so no images or keys leave the device; scoped output to visible building condition, excluding claims about neighborhoods, safety, or people.

Python, JavaScript, Chrome extension, local vision model

Architected an AI orchestration system with file-based external memory and multi-instance parallel processing, designed to preserve progress through repeated system crashes across eight hours within 2.99GB of allocated RAM.

Field-tested the design at the 2025 DoD Cyber Sentinel Challenge, where the architecture produced solutions across categories including forensics and web security, and sustained operation through repeated crashes.

Ubuntu virtual machine, LLM orchestration

Identified a gap in security tools for small organizations, which face common threats like phishing without the budget or expertise enterprise tools require; defined requirements for an accessible, privacy-first security platform.

Built a browser-based prototype with client-side OCR, real-time threat feed aggregation, and a Visual Threat Score, validating that email threat analysis can run fully client-side.

HTML, JavaScript, Tesseract.js