English entry points for 12 technical articles: architecture decisions, native GenAI in .NET, real-time systems, IoT, performance and product engineering.
Featured story
A long-form history of alternative GTA multiplayer networks, written from the perspective of someone who built and operated servers in that ecosystem for years.
What an LLM integration really costs in production - the three cost centers a POC hides, realistic timelines, the five traps that turn a brilliant demo into a money pit, and the cases where the right answer is no AI at all.
The exact day-by-day breakdown of my Product Sprint offer - from the free scoping call to the PR merged in your repo. What you receive each day, what I refuse to take into scope, and why a deliberately small perimeter ships more than an ambitious one.
A pure .NET RAG demonstrator that refuses to output an answer without a cited source.
How reactive state with ActualLab Fusion changes user experience: no polling, no manual cache invalidation, no duplicated state.
Why I removed REST calls and designed server-side anti-cheat for a hostile, moddable runtime.
The line between a senior engineer and someone who thinks they're senior: 'we commit to not doing X' vs 'doing X is technically impossible'. On my projects the critical rules don't live in a wiki - they're enforced by the compiler, the CI, the type or the domain object. Five real examples.
The 2026 default reflex for GenAI is 'a Python micro-service next door'. On most of my products I don't have it - RAG, embeddings, scoring and guardrails run in C#/.NET. Why that choice holds, and the one exception where I accept a Python sidecar.
Before .NET and GenAI I shipped industrial embedded firmware at the scale of ~150,000 systems per year - where a regression can't be fixed with a commit. The discipline that demands, and why I reapply it on critical server systems.
Four real production decisions: three paid off, one cost six weeks. The useful part is the trade-off, not a generic best practice.
Production forecasting is not 'train a model and read its predictions'. Three problems surface: no single model fits every product, a good model starts lying silently, and the right price is never in your gut. Here is how I solved them in .NET.
PromptVault's reversible Privacy Shield masks sensitive values before an LLM call and restores them afterward.
From SA-MP and MTA to FiveM, alt:V, RageMP and the Rockstar acquisition - where my V-Multi work fits.