Hi, I'm Ryan.

I’m a machine learning engineer based in New York. I’m interested in computer vision, representation learning, and learning from images and sensor data. I like designing experiments, understanding why models behave the way they do, and building the software to put ideas to the test.

I’ve adapted vision foundation models for pathology image segmentation, built workflows for molecular screening and generative design, and shipped an AI platform for medical intake.

Current things I'm exploring: "online"/"goal-oriented"-RL (like AlphaFold for protein generation, or LinkInvent for linkers), causal inference approaches in deep learning, and kernel optimization.

Other interests: bioenergetic dieting, biohacking, alpine skiing, squash, philosophy of empiricism, ricing my Linux server.

Stylized Art Deco illustration of an East Village street at dusk
Photo of the East Village I took, stylized via ControlNet with the Art Deco Posters SDXL LoRA.

Blog Posts

Profiting Off Zyn with BFS Webscrape

I developed a BFS algorithm to track the rapid growth of Zyn in the U.S., leading to a highly profitable investment in parent company Swedish Match.

Diagram of breadth-first search drawn as a tree of Zyn cans, with arrows visiting each level from left to right

The Butterfly Effect: How Random Initialization Shapes WTE Semantic Similarity

An experiment exploring how random weight initialization leads identical language models to develop divergent word token embeddings under identical training conditions.

Training and validation perplexity curves for the two models in the random initialization experiment

Links

Contact me at ryan.douglas.kish@gmail.com.