Alex Margolick, AI implementation

Systems I've built.

I design, build, and run AI systems that do real work unattended: multi-agent pipelines, a voice agent on live telephony, a website that improves itself on a schedule, and enterprise reporting automation. Four are below, each described with its stack, its cadence, and what it actually does.

Multi-agent automation

An autonomous job-search engine

A hybrid local and cloud multi-agent system that runs every 4 hours, unattended. Each cycle it sources new roles, verifies that a posting is real and still open before spending effort on it, tailors my resume to the listing, prefills the ATS application, drafts recruiter outreach, and logs every action to a CRM.

Anything that needs a human decision gets split out, staged, and emailed to me with a one-click link and an answer key. The engine does everything up to that click. It is built on Claude agents and browser automation. If I emailed you about a role, odds are this engine surfaced it first.

Voice + telephony

A production outbound voice agent

An outbound voice agent on Twilio, deployed with a live call-test API on Cloudflare Pages so a real test call can be placed against the running system, not a demo video.

Two details matter here. The speech uses tuned SSML prosody, so pacing, pauses, and emphasis sound like a person rather than a screen reader. And the conversation itself is designed from behavioral science: how it opens, how it handles hesitation, when it asks and when it listens. Voice AI fails on those two things far more often than on the model.

Self-optimizing web

margolickai.com, the site itself

I designed, wrote, built, and deployed margolickai.com on Cloudflare Pages. It serves an AI assistant and a live demo generator on Workers AI, plus a hardened lead-capture API with origin checks and consent gating.

The part I would ask about in an interview: the site maintains itself. A scheduled routine ships one improvement per run, verifies it against the live site before calling it done, and rolls back automatically if verification fails. Every change lands in a shipped log with its evidence. That is the same discipline I bring to client and employer systems.

Enterprise reporting

ESG reporting automation at scale

Before I built agents, I automated enterprise sustainability reporting. At McDonald's, I cut the ESG RFI backlog by 50% with AI-driven disclosure automation. At Holman, I built the company's first audit-ready Scope 1-3 greenhouse gas inventory on EnergyCap and Workiva.

Same instinct, earlier tools: find the repetitive knowledge work, automate the drafting, and keep the output at a standard an auditor will accept. That bar, audit-ready rather than plausible, is what I now hold AI systems to.

Evidence before "done"

Nothing ships on assertion. Every change carries a test result, render, or live check.

Built end to end

I own the whole stack on these systems: design, code, deployment, and upkeep.

Runs without me

Scheduled, guarded, and idempotent. If a run fails, it rolls back or flags me.

Hiring for AI implementation?

I will walk you through any of these systems live, on a call, including the parts that broke and how I fixed them.

[email protected] LinkedIn

Or text: (609) 314-4919