Hard InterferenceStop scrolling.
Start building.
Hard Interference — independent AI guides, build journals, and benchmarks from the real workbench. No subscriptions. No lock-in. Just computing you own.
781.5M
Cumulative Tokens
£40/mo
OAuth Cost
£9,612/mo
Opus 4.6 API Est.
£9,572/mo
Estimated Savings
What I Cover
Hard Interference — independent AI guides, build journals, and hardware know-how.
📖
AI Guides
The facts. Step-by-step guides, model comparisons, and hard numbers for running independent AI.
🔨
Build Journal
The story. Real builds, wow moments, and honest accounts of what worked and what didn't.
📈
Benchmarks
New model drops, I benchmark them for you. Real-world numbers, no marketing spin.
Latest Posts
Fresh from the workbench.
APR 13, 2026
Build Journal
Building Mission Control (Or: How I Learned to Stop Worrying and Love the Dashboard)
From a static HTML file to a full Next.js dashboard. Making token drain visible, fixing cost calculation bugs, and learning that you can't cut what you can't see.
APR 12, 2026
AI Guides
Smart Model Routing: Cloud When You Need It, Local When You Don't
Stop sending every prompt to expensive cloud models. Smart routing with character and word thresholds means simple tasks hit local models for free, while complex reasoning gets the cloud power it needs.
APR 12, 2026
AI Guides
The Discord Token Wake-Up Call
A bot token was accidentally shared in a chat session. The redaction system caught it, but it forced a hard question: how safe is prompt chat? Does information leave this PC? The day we got serious about privacy.
APR 12, 2026
Benchmarks
Choosing the Right Models (So You Don't Burn Money)
Benchmarking 6 local models on the RTX 5070 Ti. Some scored 10/10 in under a second. One scored 7/10 but took 78 seconds. The benchmark results that shaped our entire model routing strategy.
APR 11, 2026
Build Journal
Day 1: The Box Arrives
Alienware Aurora with an RTX 5070 Ti, 64GB RAM, and a dedicated Samsung 990 Evo Plus for scratch testing. Ubuntu 24 installed, Ollama pulled, and the ambition to run AI without a cloud bill. The first steps -- and the gap between 'it runs' and 'it runs well.'