25.76°N 80.19°W
I build agent workflows that do real work under human gates, and I get the teams who did not ask for them to use them. Open the booking app and book a real slot. The tests, the limits and the source are public.
available now · contract or full-time · remote or on-site · Miami, FL
459tests
pass · 0 skipped
240trained
16-seat cohorts
$20K/yr
eliminated · still running
8+yrs
business analysis
3langs
EN · ES · PT
about
I have run the rollout for people who did not ask for one, and won over the process owner who did not want the change by leading with the relationship, not the analysis.
I write the specs, the acceptance criteria and the gates. The crew writes the code. I merge every change.
Eight-plus years of business analysis across utilities, supply chain and finance operations sits under all of it.
shipped
booking_app · deployed, not a mockup
Live on my own tenant. A visitor can book a real slot. Row-level security is enabled and forced on every table. Tenant identity is set server-side, never taken from the browser.
// the numbers, and the test that tries to break them
Revert the isolation migration and 29 of 44 isolation tests fail. Drop one policy and 3 of 3 fail for that table. This is the test that would expose the isolation if it were decorative.
// status, stated plainly
wired
not yet
// I wrote the specs and acceptance criteria in Linear. The crew built it. I merged every change.
development_engine · public repo
A public repo where a crew of agents ships a Linear backlog under separate builder, reviewer, planner, researcher, spec and security seats. Two human gates: I approve direction up front and merge every change at the end.
Branch protection is enforced, and proven. A direct push from the crew's own
credentials was rejected (GH013). Routing rule:
model = max(points-tier, risk-tier).
Risk floors a task's treatment up.
151 issues audited. The agents caught 11 errors humans made or approved. I caught 4. My confident prescriptive spec details were wrong 28 times. My end-state specs survived 28 of 28.
read_the_source ↗// what holds
// where a human is required
site_pipeline
Six stages, each with its own quality gate. Discovery brief in, deployed Astro site out. This page came through it.
// 04 and 05 run in parallel
measured_time
About 88 minutes, brief to deployed. Measured once. A best case actually achieved, not an average.
Several customer sites have shipped through it. They are public on my GitHub.
adoption
"The last automation project died in the sales team's hands."
I build for augmentation, not replacement: the value of AI is not the cost you cut, it is the backlog you finally deliver and the lives you make easier, for customers and for the people serving them.
enablement · built from scratch
I ran change management and user adoption for a GCP enablement program I built from scratch at FPL. Trained 240 people in 16-seat cohorts. Stood up a community of practice that outlived the engagement. Automated the program's own workflows with Power Automate.
resistance · relationship first
A reverse-logistics routing redesign at UDT removed $5K/yr in cost. The process owner resisted. I won them over by leading with the relationship, not the analysis.
// the craft underneath
ops_layer
How I work arrives with me, already built and running. Scheduled agent routines do the checking. The boards are part of it: agents read and write Linear and Todoist directly, file their own work, and flag anything only I can do.
"I don't ask it to do the work, I ask it how the work should be done."
daily_sweep
Reconciles Gmail, Calendar and Todoist against each other, flags anything inbound with a clock on it, and corrects any board that disagrees with the inbox. Built after a missed reply cost a live opportunity.
the_vault
An Obsidian knowledge base my agents read and write mid-task, under provenance rules. Every note carries who wrote it and whether it was verified. One home per fact.
send_gate
The agents draft emails, applications and filings completely. I press send. Nothing goes out under my name without my hand.
// every one of these drafts. none of them sends. I press send. · daily_tools: Claude and Claude Code, OpenAI / ChatGPT, Perplexity, Grok
lineage · 2017 → now
2017–2019 · Wythe Capital Trust · Data Analyst
Selenium-automated extraction pipelines feeding SQL and statistical analysis for real-estate lead targeting. Salesforce data analysis and reporting only, never administration or configuration.
2019–2024 · IPC, Subway’s supply chain arm · three roles
A multi-language legal-compliance system in Excel, VBA and Power Query, documented end to end in Visio. An MDM vendor transition with a managed cutover. Automated Tableau reporting. Supporting a franchise network of 20,000-plus restaurants.
2024–2026 · UDT · Business Analyst
Invoicing automated for an entire business unit: SQL feeding Power BI reports, proven first in Excel. $20K/yr eliminated, still running without me. A $15K/yr at-risk contract rescued. The org-wide ERD, the company-wide data strategy, and the $5K/yr routing redesign.
2026 · FPL · Business Analyst II, contract
The GCP adoption enablement program above, built from scratch. Power Automate automating the program’s own workflows.
2023 → now · LLMs
LLMs since 2023. Then the agent systems: the engine, the booking platform, the pipeline, the ops layer.
skills
Claude / Claude Code · agentic system design · Power Automate · Selenium
SQL (expert) · Python · Excel (expert) · Power Query · VBA
Power BI incl. paginated · Tableau · data modeling and ERDs · MDM · data governance
requirements engineering · process mapping (Miro/Visio) · UAT · Scrum · Jira · Linear
contact
The risk is not a bad hire. It is another quarter with a demo behind you and nothing your team uses.
Send one email. Subject line: Hiring inquiry: AI Workflow & Automation.