back
user profile
mikeayles
396karma·75submissions·November 25, 2025
about
I build complex technical tools, from automated hardware development pipelines to raycasting engines in CAD software.
The DOOM on Engineering Tools Series:
OpenSCAD-DOOM: Raycasting in a parametric CAD language, now playable in browser. (The Register)
KiDoom: DOOM rendered on PCB traces. #1 on HN. (The Register, Hackaday)
ScopeDoom: Vector rendering on an oscilloscope via audio DAC. #1 on HN. (The Register, Hackaday)
The Serious Work:
Phaestus: Compiler for physical products. Generates schematics, PCBs, enclosures, and firmware from natural language.
Bitwise-MCP: MCP server for embedded systems. Indexes 1000+ page reference manuals with hybrid keyword/semantic search.
TOKN: Token-Optimised KiCad Notation. A compact encoding format that reduces schematic token count by 92%, enabling LLM circuit generation. Includes benchmark suite comparing model performance on hardware tasks.
Engine Simulator: Thermodynamic diesel engine model at 0.1° crank resolution, vanilla JS in browser.
I specialise in solving "impossible" technical problems and shipping validated products.
mike@mikeayles.com | mikeayles.com
recent activity (75 total)
comment
I 100% agree the model is basically useless. that was never the deliverable. the artifact here is the inference engine, not the model living in it. 3.16M params at character level is just what fits in…
comment
Appreciated. and yeah, agreed the interesting comparison isn't "is this scalable as-is" (it isn't, PL fabric, cost, power), it's that the CIM / analog-compute startups yo…
comment
Just a quick test to see if i'm back
comment
You're correct, the flat line is aggregate only. the fabric is saturated from a few dozen active clients onward, so extra connections can't buy throughput, they just queue. per-user p50/…
comment
I just posted a Show HN of my most recent side project, a live demo of a tiny-llm implemented in FPGA fabric, hitting an aggregate peak of 60,000tok/s, but a 'usable' model at 21,000tok…
comment
I started this about 10 weeks ago when the Taalas chatjimmy demo first did the rounds, Llama 8B baked into custom silicon, 17k tok/s for a single user. Their whole thesis is that inference is bou…
comment
I'm using the AMD (Xilinx) K26. It's a Zynq Ultrascale+, the successor to the old classic 7000's. I'm running it on the KV260 dev board, because I'm using it for another side …
comment
It will be, I'm just not happy with my blog post before making it live. The blog will have a live interactive chat and a link to the repo with the HDL. I don't think anything I did was parti…
comment
AMD could have saved their money and used their own hardware! I've got a language model doing 60k tok/s on AMD hardware already, a Xilinx Kria K26 SOM, with the weights baked into URAM/…
comment
So for people wondering if it can be used to accelerate LLM inference, sadly not. I've been trying to hit 100,000tokens/s with a 3.28m dumb model, and even this is an order of magnitude too …
comment
you may find this useful: https://phaestus.app/blog/blog0031 Edit: Forgot I also got doom running in openscad: https://www.mikeayles.com/blog/openscad-doom&#…
comment
I benchmarked Claude Code and GitHub Copilot on the same model (Haiku 4.5) with and without RAG-powered semantic search across 60 queries on a real codebase. RAG didn't make search more accurate …
comment
Rewriting the backend Bitwise Cloud, my semantic search for embedded systems docs Claude Code plugin from Python to Go. The problem was the ML dependencies. The backend uses BGE-small-en-v1.5 for embe…
comment
Fantastic comment, thanks! A review of the state of BEV's was actually going to be one of my next articles (hopefully after the additives). Are you happy for me to drop you an email to review a d…
comment
I do now: https://www.mikeayles.com/rss.xml
comment
Good spot, and gruez is right about the caption too (fixed both, thanks). The car's L/hr figure was wrong. At 45 mpg (imperial) and 70 mph cruise, a car burns ~7 L/hr, not 3. That makes…
comment
Probably the first thing to consider is the trucks have their speed calibrated periodically to ensure the accuracy of their tachographs (in the UK at least) so a truck doing 90kmph may show as 100kmph…
comment
Sorry, got mixed up there, will amend, the 60 is for +3.5t! Edit: Nope, despite the vehicles only being able to propel themselves to 90kmph, the speed limit is indeed 60mph (in England and Wales, Scot…
comment
I have done a deep dive here: https://www.mikeayles.com/blog/on-vehicle-hydrogen-generatio... Short answer, it takes more energy to generate than the energy it produces. You can …
comment
Fair point that a full TCO comparison would be more complete, and it's something I'm planning to cover in a later post. But the capital cost argument actually reinforces the conclusion rathe…
comment
Good catches, both of them. The hydrogen pathway should include grid transmission before the electrolyser, you're right. In practice it doesn't change the overall ratio much (multiplying by …