Selected work · 2026
The Work
Lead generation, inbox triage, multi-perspective research, self-critiquing decks. Every one shipped end to end — real APIs, real deploys, real error handling.
AIS-OS — a personal AI operating system
A file-based operating system for running my work through AI: onboarding rituals, a decisions log, a connections registry, and a library of skills that wrap real automations so each is one command to run.
→ Every automation on this page is one command inside it.
02System2026Outreach engine
A multi-source prospecting engine across seventeen niches and fifty states. It finds business owners, proves each one is approachable and still trading before a human ever sees it, drafts the outreach from pages it actually read, and reads its own board back so nobody is contacted twice. Ten services, all wired as direct APIs.
→ 72 verified owners for 13 cents, and it blocks the ones it should not contact.
03System2026Nonprofit lead engine
Ranks organisations by financial distress read straight off their public tax filings, on two free government datasets and no paid data vendor. Days of operating cover, multi-year deficits and payroll collapse are derived from the raw 990 fields, then every contact name it finds is confidence-graded against its source.
→ It grades what it cannot evidence and returns “unclear” instead of inventing a lead.
04Pipeline2026Clay → Claude cold-email pipeline
Sources leads through Clay, then walks each one down a chain: qualify it, read the business, find a recent signal worth mentioning, and write a cold email that earns its opening line. Fifty leads in, fifty subject lines and bodies out, each referencing something true about that specific company.
→ Fifty send-ready emails, none of them a template with the name swapped.
05Product2026AI Lead Qualifier
A lead lands in a form; a background task scores and tiers it against a written ideal-customer profile and returns a validated verdict — score, tier, reasoning, the next action, and which of budget, authority, need and timing it actually clears. The browser never touches the job queue or the keys.
→ The scoring rubric is a markdown file, so changing what counts as a good lead is an edit, not a deploy.
09Agent2026STORM — multi-perspective research briefings
Simulates six expert perspectives on a contested question, maps where they disagree, synthesizes a briefing, then peer-reviews its own work and renders it to a clean report.
→ Six expert lenses distilled into one brief you can decide from.
12Automation2026Slide Builder with a vision loop
Builds Google Slides decks from scratch, then renders each slide to an image, looks at it, critiques it, and fixes it — a self-correcting loop instead of one blind pass.
→ The deck screenshots itself and fixes what looks wrong.
- 06
Inbox → ClickUp triage engine
Collapses inbox triage into a single command.
- 07
Estimate Desk — n8n quote generator
Falls back to a locally-computed quote with the webhook unset, so the app runs with no backend at all.
- 08
Strategy report automation
Push to main and GitHub Actions redeploys the task. Nobody touches a server.
- 10
Daily research agent
Runs on a schedule and files itself — built for nobody to be watching.
- 11
AI news digest → Notion
Ships on-command instead of on a cron — a deliberate call to stop it burning credits daily.
- 13
Resume Screener
Verdicts tied to source text, with “unclear” kept visually distinct from “not met.”
- 14
mertozcetin.com
You are looking at it. The work is the only thing that gets full colour.