Research, tools and a few things I built to see whether they would work. Roughly newest first.
WMF Benchmark
Measuring how world models forget — 2026, in progress
Perseo
My longest-running project — a personal assistant that listens and watches
HybridMamba-11
31.8M params in 13.6 MB — first SSM entry in OpenAI Parameter Golf
Jetson Orin
Running an LLM in orbit, or trying to — thesis proposal
Multilingual Benchmark
Chinese vs. Western frontier models across nine languages
Armario
Outfits from the real forecast — rules I can read, no model
Netkey
NFC networking startup where I was CTO
Telefónica
NB-IoT and swarm robotics talks in Seville
Sputnik
The seminar that changed how I pick problems
AI Mouse
Turning a hand into a pointing device
Face Detection
Where I learned that a working demo is not a system
ClassTranscriber
Lectures into speaker-separated notes, all running locally
AutoLinkedIn
Retrieval-augmented drafting from my own writing — paused
Agentic Reasoning Traces
15 traces, 142 steps, three of them wrong on purpose
ByTheWay
An early venture I co-founded
OrgaAI
Conversational AI under NDA — most of it I can't tell you
