Autonomous Aerial Vision System
As the Computer Vision Lead for Triton UAS, I architected a C++ inference pipeline that powers our autonomous planes. I integrated fine-tuned YOLO v11 on a Jetson Orin Nano for real-time target detection and built a synthetic data engine that doubles as a mock camera for ArduPilot SITL simulations, allowing us to validate our full detection-to-geolocation stack in a virtual environment.
Learn More →File Order Randomizer
I identified a workflow gap in Adobe Premiere Pro and engineered a native extension to fix it. Built with Adobe CEP and ExtendScript, my tool allows editors to instantly randomize clip order inside the timeline. It handles complex bin recursion and clip timing logic, serving over 8,000 users with a 4.9-star rating on the Adobe Exchange marketplace.
Learn More →Self-Fed Memory: A "Second Brain" AI
I built a RAG-based AI assistant that acts as a “second brain” by ingesting my personal notes and conversations. Unlike generic chatbots, it uses time-weighted vector retrieval (Pinecone) and automatic preference learning to give answers grounded in my actual life context. I engineered the full stack using FastAPI, Docker, and Streamlit to create a system that “remembers” who I am.
Learn More →Disinformation Detection: Benchmarking Deep Learning Models
I engineered a benchmarking framework to evaluate deep learning architectures against human judgment in detecting fake news. I trained custom LSTMs and fine-tuned transformers (DistilBERT), achieving 99.43% accuracy on in-distribution data. The study highlighted the critical “generalization gap” in pre-LLM models when facing out-of-distribution real-world media.
Learn More →