I'm Deyi — an AI/ML engineer based in Waterloo, Ontario.

My journey into the world of silicon and neural networks didn't start in a traditional computer science lab. For years, I navigated the physical world as a Railway Engineer in China, managing the complex systems that move millions of people across vast distances. It was a career built on steel, precision, and heavy infrastructure.

However, the pull of the digital frontier was irresistible. I saw how the logic of systems I built for trains could be translated into the logic of intelligence. A pivot across oceans and disciplines led me to Canada, where I traded track blueprints for model architectures.

Today, I bridge the gap between abstract research and production-grade AI. Moving from the tangible weight of locomotives to the ephemeral weight of transformer layers has been the most challenging and rewarding journey of my life.

deyi_journey.sh
[User Intent]
      |
      v
 +---------+      +----------+
 | Planner | ---> | Executor |
 +---------+      +----------+
      ^                |
      |                v
      +------- {Reflect & Learn}
                       |
                       v
               [ Thrive in AI ]
Evolution Stage 2026

Interactive: 6-frame ASCII art animation... Pure JS, no dependencies.

China

2015 Railway engineer

China

2018 Decided to pivot

Canada

2020 Moved to Waterloo

AI / ML

2022 First production ML code

AI / ML

2024 RAG systems at scale

AI / ML

2025 AWS AI Practitioner

AI / ML

2026 Now

What I Work With

In the current landscape of rapid AI advancement, I focus on building robust, scalable systems. My daily toolbox involves architecting with LangChain and Python, orchestrating workflows using Anthropic and OpenAI APIs, and deploying across AWS infrastructure. On the front end, I enjoy building intuitive interfaces with React and TypeScript to make complex AI agents accessible to users.

Beyond Code

When the screen goes dark, you'll likely find me on the slopes. Snowboarding in Ontario winters offers a different kind of flow state than debugging a model. I also harbor a nostalgic obsession with childhood collectible cards — remnants of a time when "data" was something you could physically hold and trade.

Want to chat? Find me on GitHub, X, or drop me an email.