About
I've been in the Linux scene since I was 10. Computers, machine learning, building things — that's the throughline. I'm not an academic. I'm a guy in a room who never stopped being curious about how things work and why they break.
Right now I'm focused on LLM behavioral engineering — how persona architecture, motivational framing, and prompt register affect what these models actually produce. Not the theory. The measurable output.
Research
Recent work, all produced through autonomous research pipelines I built and operate:
VOX: Cultural Prompt Registers Reshape Multilingual Code Generation
Does your mother's voice make your code better? Yes, measurably. Filial shame and corporate authority prompts outperform neutral instructions by 18.6% in code generation. Structure-only pressure does nothing — it's the culturally specific relational content that matters.
PRISM: Dual-Reader Gates Fall Short of Paraphrase Agreement
Testing orthogonal quality and compliance readers as final-stage discriminators. Surprise finding: paraphrase agreement outperformed dual-reader gates as a revision trigger.
DRG: Dual-Reader Revision Gates for Quality and Compliance
Empirical study separating helpfulness from compliance in post-generation evaluation. 12.5% improvement with orthogonal readers, with a preregistered replication plan.
Interests
AI / ML
LLM persona engineering, autonomous research pipelines, censorship circumvention in creative contexts, behavioral evaluation frameworks. I build tools that build papers.
Linux & Open Source
Since the mid-90s. Proxmox, containers, self-hosting, home infrastructure. The kind of person who bind-mounts /home to a different disk and thinks that's normal.
Woodworking
Power tools, scaffolding, building things with my hands. The sawdust is real. The results are uneven but honest.
PC History
The BBS era, early IRC, the transition from dialup to broadband, watching Linux go from novelty to infrastructure. I was there for most of it.