I design and build AI automation that survives real use: research platforms, opportunity pipelines, document intelligence at drive scale, and multi-agent orchestration — optimized for two hours of operator effort per day.
Chapter 01 — The problem
Every AI-generated page looks the same. Every automation breaks when reality shifts. The gap between "it works on my machine" and "it runs itself for months" is where real engineering happens.
LLMs produce template output unless forced through hard gates. My pipeline runs 57+ slop checks before anything ships.
Most "opportunity finders" violate platform ToS. I build on official APIs or walk away from the source entirely.
An automation that needs daily babysitting failed its purpose. Everything I ship targets ~2 hours/week of operator time.
Chapter 02 — The journey
Each one solves a different failure mode. All are in production today, measured by how little attention they need — not by how impressive they look in a demo.
Five classical tafsir volumes processed into structured data with word-level morphology, lexicon entries, and embedding-based verse search. Ships as an automated daily Telegram bot delivering layered breakdowns at 04:00.
Brand-DNA extraction from client's own assets (menu PDFs, Instagram grids) fed through a 57-gate anti-slop filter and macrostructure framework. Mobile verified via CDP screenshot loops at every breakpoint.
Multi-source lead tracker with effort-adjusted ROI scoring: skill match plus hourly-rate estimation minus sync-meeting penalties. Official Upwork GraphQL, never scraping. Human picks, system prepares.
Dual-account IMAP/SMTP automation across personal Gmail and corporate Zimbra. Credentials stored locally with strict permissions — they never touch chat or cloud.
Five specialized AI assistant profiles with cross-profile shared memory (observe/claim/compile pattern). Ten-plus scheduled jobs drift-guarded against provider changes. Each profile has its own domain, memory, and voice.
Chapter 03 — The proof
Not projections. These are current counts from running systems.
Closing
This page itself follows the same principle — static HTML, no tracking, no build step, no dependencies beyond two Google Fonts. It will still render correctly years from now. That's the standard.