I am a Machine Learning & Computer Vision researcher focused on solving difficult, unstructured data problems. I specialize in 2D/3D segmentation, Document AI, and LLM/VLM fine-tuning.
I don't just train models; I build data engines and custom architectures to extract signal from extreme noise.
I won the First Title Prize by decoding the unreadable. My team developed a custom 3D Computer Vision pipeline to detect ink in carbonized Herculaneum scrolls.
- The Problem: Segmenting characters in 3D volumetric X-ray scans where the human eye sees nothing.
- The Solution: Designed MiniUNETR (a lightweight 3D Transformer/CNN hybrid) and built an iterative "ignore-mask" data engine to handle uncertainty.
- The Result: Successfully extracted the first readable title from Scroll 5.
- View the Winning Code & Case Study
|
|