Blog

Subscribe via RSS

August 2026

On-Chip LLM: Sampling Without Asking

The FPGA decoded a reply in 6.6ms. It landed in the browser in 100. The gap was one unmeasured loop, and the fix was the Gumbel-max trick.

On-Chip LLM: Inside the Chip

The datapath of a 3.16M-parameter transformer living in FPGA fabric: wide-word GEMV, the dual-port split-brain, and the proof that 16 streams is a hard wall.

On-Chip LLM: Serving an FPGA to Strangers

Browser to transistor: a Cloudflare tunnel, a serving box, a Tailscale link, and the trick that lands a reply the instant you press Enter.

On-Chip LLM: War Stories and the Method

Bit-honest before fast. Where iverilog lies, why silicon beats the timing report, and the optimisation that won and got stopped anyway.

June 2026

May 2026

Banger: A Local Photo Culler That Doesn't Touch the Cloud

A files-only photo culler for the laptop. CLIP embeddings, MMR diversity, ArcFace identity clustering, and a Ridge regressor trained on your own taste. Copies keepers out, never edits the originals.

April 2026

Does RAG Actually Help AI Coding Tools?

I benchmarked Claude Code and GitHub Copilot on the same model with and without RAG-powered semantic search across 60 queries. Both tools search equally well — but RAG cuts token consumption by 23% and helps on the hardest queries.

March 2026

What's Actually in a Litre of Diesel

Brent crude hit $119.50, then crashed to $88, then bounced to $90, all in 48 hours. Your fuel bill moved 5p. The anatomy of a diesel price, why wars move it less than you'd think, and why the Chancellor is still a bigger threat than Iran.

It's Just Weights

A tiny neural network reproduces the Cortical Labs CL1 DOOM demo in your browser. 132 parameters vs 200,000 biological neurons.

goformersearch: Vector Search in Pure Go Featured

Brute-force and HNSW nearest-neighbour search with zero dependencies. When to use which, backed by data.

February 2026

The Hydrogen Truck Problem Isn't the Truck

Hydrogen fuel cell trucks exist. They work. The problem is everything around them: production efficiency, infrastructure economics, and a competitor called 'the grid' that's already everywhere.

50kWh In, 33kWh Out: On-Vehicle Hydrogen Generation

The energy chain behind HHO systems, worked from a real product's published specifications. Why on-vehicle hydrogen generation for diesel trucks is a thermodynamic impossibility, not an engineering challenge.

January 2026

Playing DOOM in OpenSCAD at 10-20 FPS Featured

DOOM in OpenSCAD at 10-20 FPS - real-time 3D rendering in a parametric CAD tool using the Manifold geometry kernel

Zookeeper: Pokédex for Real Zoo Animals Featured

Pokédex for real zoo animals - AI-powered collection game that turns zoo visits into adventures with Gemini vision identification

Phaestus: Compiler for Physical Products Featured

Compiler for physical products - natural language to manufacturable PCB, enclosure, and firmware using constrained AI generation

December 2025

TOKN: Token-Optimised KiCad Notation

Token-Optimised KiCad Notation - 92% token reduction format for LLM circuit generation, with benchmark suite and model comparison

November 2025

KiDoom: Running DOOM on PCB Traces Featured

Running DOOM on PCB traces - vector-based rendering of the classic 1993 shooter using KiCad's PCB editor at 10-25 FPS

KiCad Netlist Tool

Tool for reducing LLM token count from Kicad SCH files, to allow for automated documentation

Supplement Stack Saver: Budget-Friendly Alternatives

The supplement industry thrives on branding and marketing, often charging premium prices for common vitamins and minerals. Supplement Stack Saver cuts through the marketing to reveal a simple truth...

TheDuck.chat Featured

AI-powered chat application. Developed for the T3 Cloneathon. DB currently inactive, demos available in repo.