Here is how I think and what I build.
I am a non-linear systems builder. I work in high-velocity bursts and produce extensive, complex infrastructure. This is not a resume. This is proof of work.
The three service sites on this domain say hire me to do this task. This one says something different: here is what I actually am.
12-Pipeline AI Work Infrastructure
Built a full AI-assisted personal operating system from zero in 22 days. Twelve named pipelines with individual A/B state file rotation, cognitive layering across three document tiers (Active → Log → MASTER distillation), and zero context collapse across 34+ sessions.
- A/B state rotation across all 12 pipelines — no session starts cold
- Cognitive system: 3-layer architecture separating working state, session log, and distilled master
- Accomplishments tracking with the same 3-layer merge protocol
- Automated system health check running via Cowork scheduled task
- System inventory auto-generated: 372 files, 180.5 MB as of April 12, 2026
- SysOps automation: folder cleanup (SHA-256 exact + rapidfuzz near-duplicate detection), PDF OCR rescue pipeline, ChromaDB knowledge base (44 MB) with ingest and query scripts
This infrastructure is the environment in which everything else on this page was produced.
AH Console v1
Built a custom Python/PySide6 desktop HUD with a Giger biomechanical skin from zero in a single 13-hour session on April 6, 2026. Full daemon architecture with a 2-minute router cycling between live API integrations.
- Stack: Python, PySide6/Qt6, custom CSS-equivalent skin layer
- Live integrations: weather, moon phase, Gmail unread count, RSS feeds
- Daemon/router architecture: modular, swappable data panels
- Three wireframe iterations (terminal skin, minimal skin, final Giger design) before build
- Zero prior Qt6 experience at session start
The 13-hour sprint is not the exception. It is the standard operating pattern.
VoiceArchive & Recursive Agent Tuning
Encoded authentic human voice from 99,262 Facebook messages spanning 15+ years into a queryable AI context system. Built the extraction pipeline, the encoding methodology, and the query architecture from scratch.
extract_fb_messages.py— 450 MB Facebook export processed down to 9.4 MB clean text (318 threads, 99,262 messages)- ChromaDB knowledge base: 44 MB vectorized voice store with ingest and query scripts
- Three VoiceArchive tiers: Creative, Everyday, and MASTER distillation
- Circuit breaker logic built for agentic feedback loops: semantic clustering, ambiguity detection, human-in-the-loop override pattern
- Lyric-deep-then-compress workflow formalized as a pipeline-agnostic writing pass variant
- 86.7% Prolific approval rate across 98 submissions — voice encoding producing measurable output quality across platform AI screening
The agent tuning problem is: how do you give a system enough context to stop producing generic output? This is my answer to that question.
Full Python Stack
- Horned Angel Studio — Python/tkinter desktop app, 4 of 5 tabs completed (~2,452 lines in Photo Studio module alone)
- prc_notifier — monitoring agent for review platform slot availability
- folder_cleanup.py — SHA-256 exact + rapidfuzz near-duplicate file detection and cleanup
- PDF OCR rescue pipeline — pdftoppm + Tesseract batch processing
- Image asset generation pipeline — Playwright + Photopea headless automation
- 7 websites built and live — full PHP contact forms, custom SVG headers, gig collateral PDFs, sitemaps
- 1 digital product live — Oblique Doorways Pack 1 on Gumroad at $17.99
Reach Out
I do not participate in traditional multi-round interview cycles. If you are a technical lead or founder working on a complex agentic or infrastructure problem, I am available for 5-day paid technical trials.
Send three sentences: who you are, what you are building, and what the problem is. I will tell you whether I can help.