About
I'm Steven W. White. This site is named for Key Largo—the first island on A1A, and where many of my journeys begin.
Somewhere between the noise and the sea, there's a place where the thinking clears. No deadlines pressing. No meetings waiting. Just the opportunity in front of me and the time to think it through. I write from that place whenever I can find it.
See my favorite spots along the wayThe Journey
25 years building things—first telecom networks, then neural networks. In 2009, I founded ImageVision, a computer vision company building image recognition and content moderation for social platforms. We raised $12.5M, landed contracts with Google, Facebook, Apple, Samsung, and In-Q-Tel, and built systems processing billions of images. Two Hat Security acquired us in 2019, and Microsoft acquired Two Hat in 2021. That technology now helps moderate Xbox, Minecraft, and MSN.
After the exit, I joined NVIDIA as a Deep Learning Solution Architect—TensorRT optimization, edge inference, industrial IoT. I joined AWS in 2019, working as an AI/ML Solution Architect and later building security AI systems with multi-agent architectures.
After seven years at AWS, I moved to Google, where I work today. It's the next chapter in my journey through AI and software engineering.
Two decades of lessons:
- Demos and deployment are different planets
- The model is 10% of the work
- Latency budgets teach you what matters
- Ship it, then improve it
Building FoilFox
Outside my day job, I'm building FoilFox, an app for trading card collectors. It brings computer vision to the cards in your hands: scan a card, check its identity and finish, and organize it in a digital binder.
Working on FoilFox brings me back to the details that make computer vision useful: lighting, latency, ambiguous matches, and helping someone get the right result. It's another way I learn by building.
What's Here
These tutorials are what I wish someone had handed me when I started. Every one has complete, runnable code—if you can't run it and learn from the output, I haven't done my job.
- Embeddings — Bi-encoders, cross-encoders, semantic search
- Retrieval Systems — RAG, ranking, vector search
- Agents — Tool use, orchestration, real-world APIs
Background
- M.S. Data Science — University of Texas at Austin
- B.S. Computer Science — University of Texas at Dallas
- 2 patents in computer vision and machine learning
Connect
By day, I work at Google. This site reflects my personal views and projects, not those of my employer.