Apr 20, 2026/Tech
Linear Attention Explained: The Core Concept and the Math, Without the Fluff
This article cuts the fluff, gets straight to the core concept and math, and explains every variable as it appears. Nothing is left undefined.
~/nathan
building practical AI systems
session://blog/tech
$ find blog/tech -type f
### Tech Writings
Systems, ML, LLMs, and the engineering details that decide whether ideas survive reality.
13 tech articles
Open article
Apr 19, 2026/Tech
Sliding Window Attention Explained: The Core Concept and the Math, without any fluff :)
This article is based on Longformer: The Long-Document Transformer (Beltagy, Peters & Cohan, 2020, arXiv:2004.05150). The paper introduces a scalable attention mechanism for long documents;...
Mar 26, 2026/Tech
Why Crypto Has Not Beaten Wall Street Yet (And Why That Question Might Be Wrong)
TL;DR: On public blockchains, your transaction sits in a visible waiting room before it goes through. Automated bots and block producers can read it, act on it, and leave you with a worse...
Mar 24, 2026/Tech
vLLM: A More Efficient Way to Serve Large Language Models
When an LLM generates a response, it stores intermediate computations called the KV cache in GPU memory at every step. Traditional serving frameworks pre-allocated memory for each request...
Mar 19, 2026/Tech
Git Worktrees: The Essential Git Feature Many Developers Still Do Not Use
But there is a feature that quietly solves one of the most common sources of friction in everyday development. Many developers have heard about it. Very few actually use it in practice.
Mar 18, 2026/Tech
Google Analytics Worked on Localhost but Failed on GitHub Pages. Here’s What I Learned.
A beginner-friendly debugging story about static sites, Next.js, Google Analytics, and why production can behave very differently from local development.
Feb 20, 2026/Part 7
Part 7: Putting It All Together (What This Actually Means)
But I’m starting to see the path forward more clearly. Let me share what I’m taking away from all this research and learning.
Feb 20, 2026/Part 6
Part 6: Evals, Optimization and Organizational Reality (What Actually Determines Success)
You can build the perfect model. Implement flawless pipelines. Have bulletproof monitoring. Deploy with zero downtime.
Feb 18, 2026/Part 5
Part 5: The Critical Pieces: Observability, Agents, Security
The unglamorous stuff. The debugging at 2am stuff. The “why is the model doing this?” stuff. The security vulnerabilities stuff. The “future me will hate present me for not documenting...
Feb 17, 2026/Part 4
Part 4: Production Systems (Where Good Models Can Die)
I had built a model. It worked beautifully on my laptop. 95% accuracy on my test set. Clean code. Fast inference.
Feb 12, 2026/Part 3
Part 3: LLMs and Modern ML (The New Fundamentals)
I mean, I could explain transformers. I knew about attention mechanisms. I had read the “Attention Is All You Need” paper (okay, skimmed it).
Feb 6, 2026/Part 2
Part 2: Training and Loss Functions (What You’re Actually Optimizing)
Your loss function isn’t just a number that needs to go down. It’s a precise mathematical statement about what you care about and your model will give you exactly what you asked for, even...
Feb 5, 2026/Part 1
What it actually takes to be a strong AI/ML engineer in 2026
So I’m starting a 7-part series this week to break down the fundamentals that actually matter.