Shawn NgAI engineer · writer · maker
Hey, I'm Shawn.
I talk about AI and practical ways to become more efficient, focused, and productive!
By day I ship machine learning systems to production. Here I write deep dives on the computer science nobody looks at, and share the AI prompts I use to get more done with members.
Also on YouTube: Divide and Quantum Best of the Best in AI
🧠
Deep dives that respect you
The ranking model deciding what you watch, the math holding up the internet: real systems, real numbers, no hand-waving.
⚡
The Vault
A growing library of battle-tested AI prompts for planning, writing, coding, and killing busywork. Less busywork, more real work.
🎥
Videos, twice over
Two YouTube channels: visual CS deep dives, and a running field guide to what actually matters in AI.
The Vault
Steal the exact prompts I use to work faster.
Not one-liners: complete systems with a role, a process, and an output format. Paste one into your AI assistant and it just works.
- 24 battle-tested prompts, organized by job
- Planning, email, meetings, learning, writing, and code
- Works with Claude, ChatGPT, or any capable assistant
- New prompts whenever I find something that works
No credit card. Just an email and a password.
Latest deep dives
Stories about the hidden computer science running the world.
The State of AI: A Field Report From Someone Who Actually Ships This Stuff
Halfway through 2026, the models are genuinely remarkable and the production reality is genuinely messy. A working engineer's account of what changed, what plateaued, and where the real work moved.
Local Models vs Frontier APIs: An Honest Accounting
What it actually takes to run LLMs on your own hardware, why memory bandwidth is the real bottleneck, and when the economics flip in favor of the API.
The Old Guard Refuses to Die: Classical Optimization vs the Learned Upstarts
Neural networks have been promising to replace exact optimization solvers for a decade. The solvers keep winning where it counts. Here is an honest scoreboard of classical methods versus learning-to-optimize, and the hybrid pattern that is actually emerging.
How Machines Learned to See
For forty years, computer vision meant hand-crafting features by hand and hoping. Then in 2012 a neural network cut the ImageNet error rate nearly in half, and everything we thought we knew about seeing turned out to be learnable. Here is the whole arc, from SIFT to vision transformers.
What I'm building
Products I'm shipping right now: the same systems thinking from the blog, aimed at real problems.
Krevaya
Autopilot for tech social content: trending topics in, drafts written in a voice tailor-made for each user. You approve, it publishes to your pages.
krevaya.com →
vidfora
YouTube outlier research: find the videos massively overperforming their channel and reverse-engineer why they won.
vidfora.com →