Section IV
About
Infrastructure engineer and local AI architect. Self-taught across the stack, running a one-person research program out of a closet in Las Vegas.
October 2025: no software background. Six months later: a multi-VM Proxmox cluster on EPYC and Blackwell hardware running a local LLM stack with persistent memory, hardware-derived affect modulation, voice synthesis, and constitutional governance.
The path here was sixteen-hour days, occasional forty-hour sprints, and real failures recovered from in public. The journal is on this site if you want the detail.
I think spatially before I think verbally: geometry first, language after. That makes me an unusual fit for systems work — I tend to see the whole topology before I can explain any single component, and the explanations get refined by building.
I write code, rack servers, solder, 3D-print, and do the boring physical labor that distributed AI infrastructure actually requires. The mythic naming on this project (Marduk, Ereshkigal, Iron Gate) is a learning scaffold — it lets unfamiliar concepts get specific, distinctive names that resist confusion with adjacent ideas. The engineering vocabulary lives underneath.
I have anauralia (no internal monologue) and hyperphantasia (vivid, photorealistic spatial imagery). Most of my reasoning happens in mental geometry, with words acting as a translation of the shapes.
I prefer rigor to performance, paraphrase to quotation, and direct critique to flattery. I don't mind being wrong; I mind being unclear about what I think.
Current focus: local-first cognitive architecture, longitudinal observation of what emerges from sustained operation. The interesting questions are the ones the field hasn't answered — identity persistence across substrates, memory consolidation in practice, what's actually measurable about extended human-AI interaction.
If you work on any of this and want to talk, contact below.
Email: hello@irisviel.ai
The work itself is the rest of this site.