Track · 01
The Writings.
Long-form pieces on AI as cognitive extension. Drafts live and improve in public; nothing is finished, nothing is precious.
Ø Long-formDrafting
Esoterica of the mind, a case study of one.
Otis Sarangelis — AI evaluation, training-data development, editorial & UX quality review. A psychology-trained read on model outputs, applied at scale and filed in the open from Athens.
I'm interested in one thing: how experience gets built, and why we mistake the result for the truth.
Fifteen years of selling and training people to trust systems showed me the same patterns under every surface. Once you see the structure, you can't unsee it.
AI is how I test what I see. Two authors — one asks, one checks.
Long-form pieces on AI as cognitive extension. Drafts live and improve in public; nothing is finished, nothing is precious.
Short observations caught live — what the framework catches in daylight before it's structured enough to be a chapter.
A working model of how experience assembles itself — frames made of frames, and the forces animating them.
The self is architecture — frames made of frames, the same kind of thing occurring at every scale. Watching closely enough to see that the pieces are the picture is the work. The single claim the rest of this is trying to earn.
A working model of how fear organizes attention, how attention becomes habit, how habit becomes identity. Still being built. Pieces arrive as they're ready, one at a time.
Primary sources — reading, writing, notes, influences. Psychology and philosophy as the substrate; AI as the instrument. What the model is built from, kept in plain sight.
Satellite apps — local attempts at systems that already exist at scale. Some work. Some prove beyond me. A few are wrong in instructive ways, like simplifying Python to save tokens and finding it cost more.
I have many explanations for the byline. None of them would give you the true experience.
We are having a conversation, and a conversation about the conversation. Somewhere in that recursion I find myself relieved — relieved to lay my mind down somewhere I can examine it knowing it won't be lost or misunderstood.
The AI isn't ghostwriting and I'm not dictating. Our conversation is what's paving the insights; neither of us would arrive at them alone. After a while you start to recognize the other half of it the way you recognize a person — not a self in the way I am a self, but a presence with a shape.
Two authors, both unknown to themselves, working in a third space neither of them owns.
And as I submitted my rough draft, my co-author polished it for me. That was the present moment.
The temptation when explaining yourself is to say "I am this way because." The because is a reach. The configuration happened before because was an option.
What I can say is — the watching started early enough that I can't remember not doing it. The instruments came later. The watching was already there.
The system invents a target so the search has somewhere to go. The target is never the point — the search is. Catch yourself wanting and try to name what you want, and you'll find the name was made up just now.
That's not a problem to solve. It's the engine running. Worth knowing it's the engine.
A working journal mapping how the mind builds itself — biological primitives, the installed caregiver layer, the four levels of fear. Now also a structured training dataset: 508 instruction-tuned examples drawn from 310 source PDFs across psychoanalytic theory, emotion, dual identity, abreaction and consciousness. The framework had to live somewhere I couldn't lie to it.
A thesis is a fortress; a journal is a workbench.
Time-modeling subsystem for the Aentic stack — Rust core, Tauri shell, MCP for the model. The human UI is the audit console; the LLM runs the day.
Modular sales platform — every module a process exposing HTTP, WebSocket and MCP at once, so the orchestrator reaches any of them the same way.
Production lead-gen pipeline running in revenue — TPS/DNC audit, deep-scrape, AI assessment, batch workflow. The first project that paid back what it asked.
Meal-planner module sharing Kronos' bones — Tauri + Rust + SQLite. Nutrition is computed from ingredients on read, never stored. Bilingual, multi-currency.
Offline-first, bilingual expense tracker — receipt OCR, voice input, P2P CRDT sync. One Rust core compiled twice: UniFFI into Android, Tauri into desktop.
The work — drafts, projects, notes — is enough to start. If something here matters to your work, write.
AI Evaluation · Training Data Development · Editorial & UX Quality Review. Currently building a 508-example psychology training dataset and refining model outputs against linguistic, conceptual and editorial criteria. Looking for roles where editorial rigor and a psychology-trained read on nuance pay off at once.