Independent research into how AI systems actually work. Published here.
I run experiments, build classifiers, and chase down the results that don't make sense. Most of what I find has been found before. Most of what I build exists somewhere in a better form. I document it anyway — the methodology, the dead ends, and the parts that didn't make it into the abstract.
- 01
GRAFT: Catching Grafted Code with a Compression Model
GRAFT trains a byte-level compression model on your repo's own history and flags PR changes that don't compress like your code.
- 02
Argus NVR: One Open Port, Local Face Recognition, No Cloud Dependency
How Argus NVR proxies WebRTC camera traffic through a single port and keeps face recognition, credentials, and footage entirely on-box.
- 03
Architecture Erodes One Import at a Time
drift-scan parses TypeScript, Python, and Go imports against declared zone rules and ranks every violation by blast radius.
- 04
Brace: Application-Layer Security Testing for AI Endpoints
Brace scans AI applications for the gap between what a model will do when injected and what the application does to stop it.
- 05
Compression Is the Detector: How ByteFlow Spots Malware Without Parsing a Single Protocol
ByteFlow reads raw PCAP flow bytes and scores anomalies by how hard they are to compress, with no protocol parser in the loop.
By day: making wooden assholes for hobby horses — things nobody asked for but someone had to build. By night: roughly the same thing, except the horses are language models and the assholes are benchmarks. Anonymous on purpose. No clout, no credentials — just the work and whether it's any good, done the way I find interesting.