How a simple question while playing Catan - could I generate balanced boards automatically? - became a production-grade generator, built over a weekend using an AI vibe-coding workflow with Claude Code.

A hands-on Dockerized backend platform with a production-ready microservices architecture - two independent Express/MongoDB services behind an Nginx reverse proxy, with clear service boundaries, isolated networks, and environment parity from development to deployment.

We created GUNSTANCE, the first publicly available machine-learning dataset focused purely on the "banning guns" and "regulating guns" sides of the debate, plus a hybrid self-training + LLM method that reads brand-new conversations as they unfold.

When there are more equations than unknowns, there is usually no exact solution. The method of least squares - introduced by Adrien-Marie Legendre in 1805 while studying the orbits of comets - tells us how to find the best solution by minimizing the total squared error.

A tiny GPU monitoring web app built with Django and gpustat that shows real-time GPU usage, memory, and active processes per user - handy in shared research labs and clusters.

GANs can generate useful synthetic data, and differential privacy adds a mathematical guarantee that no individual leaks from the training set. Together they enable collaborative, privacy-preserving deep learning - here is how.

How to compute the nth Fibonacci number efficiently using tail recursion in Standard ML and Python - reducing exponential time complexity to linear and stack space from linear to constant.

Solving the maximum (contiguous) subsequence sum problem in Python from a slow brute-force O(n^3) approach down to an elegant linear-time algorithm that also returns the subsequence itself.

How I went from fearing Vim to making it my daily driver for a full year, plus a practical guide to the modes, the key commands, customization, and the gotchas that trip up every newcomer.