Portfolio
AI Adoption & Enablement
I take AI from a purchased license to daily production use. At ZipRecruiter I own the revenue organization's enterprise AI program: 28 production agents, a stalled rollout taken from 2 weekly users to 88 of 112 seats in 60 days, and the governance that keeps all of it auditable. Before that, two years running my own AI-automation practice.
Seattle, WA / ZipRecruiter / 14 years, U.S. Air Force / MBA, University of Washington
Selected Projects
Re-launching a stalled Glean rollout and running a fleet of 28 production agents for a revenue org
The rollout I inherited had 2 weekly active users. People had stopped using it for two reasons: the tool didn't know how ZipRecruiter worked, and the agents were wrong often enough that nobody trusted them. I fixed the causes before touching adoption. Salesforce schema hints gave Glean the meaning of our fields; 12 skills taught it our org structure, reporting lines, enterprise segments and acronyms; the most-used agents were rebuilt first. Then the habit work: an onboarding video sent with every license that makes Glean the browser default, one-on-one sessions mapping each person's day to specific uses, and a 14-person champion network that meets weekly. Sixty days later, 88 of 112 seats were active. I specified and deployed 28 production agents across sales, trust & safety and customer success, cut the most-used one's processing time 65%, closed a Salesforce data exposure path, and built the ROI case for renewal on direct revenue attribution, the standard the EVP of Sales set.
Multi-agent Claude Code pipeline running 24/7 on a VPS, with human approval in Slack
My own channel, on retirement income rather than net worth. I specified, deployed and operate the production system behind it. A persistent Claude Code agent on a 24/7 VPS runs the whole script pipeline, orchestrating specialized subagents with structured logging, error handling and monitoring. Each week it pulls the top-performing videos from the channel and its competitors, extracts the patterns, reads comment sentiment for what's resonating or missing, researches the topic, and layers the channel's worldview on top. It drafts an outline and pauses, pushing it to Slack for approval or redirection in plain English ("less hype, lean into the tax angle"). Once approved it writes the script to the channel's tone and story patterns from a detailed CLAUDE.md. A hook agent generates 100 variations, scores them against a rubric, then ratchets the best with single-word mutations. Two review agents check the finished script, one for guideline compliance and one for factual accuracy and math, before I do the final read. Every episode also ships as a podcast and a blog post. Weekly, the system reads transcripts and performance data and tunes its own rubrics.
n8n pipeline that did what Clay does, built on each client's own stack, for 12 B2B clients
Through Inbox Effect, my cold-email agency, I specified, deployed and operated outbound infrastructure for 12 B2B clients across SaaS, agencies, events and professional services. The core was an n8n pipeline that deep-enriched every prospect on the client's own infrastructure. A research agent node pulls company news, website content and profile data into a context packet per lead; a writing node turns it into a one-line icebreaker, a tailored body and a subject line. Per-lead research went from 15 minutes of manual work to a 30-second agent run. Client systems were integrated end to end via REST APIs and webhooks (Salesforce, Airtable, Apollo, LinkedIn Sales Navigator), and I owned the deliverability stack (SPF/DKIM/DMARC, warmup, bounce tracking) through Instantly.
Full-stack no-code platform, designed and built hands-on in Bubble.io for a flat-fee brokerage
A flat-fee brokerage was running hundreds of simultaneous transactions on spreadsheets, email threads and manual status tracking. Deals got lost, documents scattered, and there was no single view of operations. I designed and built the full platform on Bubble.io, by hand and before I used any AI tooling: database architecture, role-based auth, frontend and UX. It replaced the manual workflow entirely: listing intake, state-specific compliance tracking, task management, agent assignment, escrow coordination, payment processing and revenue reporting. Stripe for payments, DocuSign for e-signatures, Zillow for listing data, OpenAI for listing descriptions, Make.com for the surrounding automation. Scoped to launch in six weeks; in the first 18 months, 1,900+ homeowners listed through it across four states, and I maintained it solo through 2026.
Background
14 years as an officer, leading laboratory operations across four bases with teams of 23 to 90 and budgets to $4M, taught me to break complex operations into repeatable processes. Adoption work is the same discipline: find the workflow, find where it fails, build the system that removes both.
I started as a medical laboratory officer running statistical QC programs: regression analysis, control charts, and HL7 pipelines routing 742K+ test results a year into Cerner under four accreditation bodies. That's where I learned to build systems that monitor their own accuracy. It's why every AI workflow I ship has structured logging, review agents and feedback loops instead of hope.
University of Washington MBA. I can sit with a VP of Sales, understand the workflow, and come back with the automation and the adoption plan that fix it. The bridge is the hard part, not the tooling.
Two years running my own practice taught me to find the problem, scope it, ship it and keep it running with nobody else to hand it to. ZipRecruiter is where that habit meets an org of hundreds: the same ownership, now with governance, champions and an executive who wants the ROI in revenue.
I'm interested in remote AI adoption, enablement and transformation roles, where the job is getting AI used, governed and paying for itself.