Nine years in fintech operations. Four years shipping product. Daily orchestrating AI workflows that compound both.
I orchestrate AI as a senior teammate, not a magic button. My day-to-day runs on Claude Code with a stack of custom skills, persistent memory, and structured plans built for the kinds of work I actually ship.
01
Claude Code for the IDE-native loop. Custom skills loaded per session (design taste, animation, polish, accessibility, performance). MCP servers connect Supabase, Resend, design files, and live documentation. A persistent memory system holds preferences and project context across conversations. Hooks enforce house rules. Subagents fan out to handle parallel research.
Claude Code · Anthropic SDK · MCP · Custom skills · Memory · Hooks · Subagents
02
Every project I lead has a CLAUDE.md, a playbook, and a memory file before it has a sprint plan. I work in plan mode for non-trivial tasks: structured spec, then implementation, then verification. Future-me and AI collaborators read the same documents. Both ship faster as a result.
Plan mode · CLAUDE.md · Memory schema · Verification before completion · Subagent orchestration
03
Seven rules of thumb that shape how I scope and ship. None invented; each one earned the hard way.
Every project I build starts with a businesses table and row-level security. Not because I plan to sell SaaS, but because the architectural discipline catches scope creep earlier.
Nine years of underwriting taught me the cheapest bug to fix is the one you scope out before building. I bring that filter to every product call.
I direct AI with structure (custom skills, persistent memory, plan mode), not autopilot. The leverage compounds when you treat it as a senior teammate, not a magic button.
Every project I run has a CLAUDE.md, a playbook, and a memory file before it has a sprint plan. Future-me and AI collaborators read the same documents. Both ship faster.
The hardest call in product is what NOT to build. I default to fewer features done well. Mocks gated to a minimum tier unless budget justifies more.
If I can’t defend a number, I don’t say it. Ranges over precision when the data is fuzzy.
I ship on Supabase, Vercel, and Postgres because they get out of the way. The hours saved on infra go into the parts users feel.
2024 / present · Founder, full-stack
Multi-tenant SaaS for barbershops in Costa Rica. Public booking, live queue tracking, owner dashboard, and PWA. One platform, three products.
2026 · Discovery lead, design direction, solution architect
Discovery, editorial design direction, and proposed architecture for a 50-year-old professional college modernizing how it serves its members. Eleven mock screens; proposal under board review.
2026 · Discovery lead, product strategist, solution architect
Scope discovery, multi-role PWA architecture, and gated mocks for a regional mobility operator unifying fleet operations, B2B contracts, and B2C ticketing. Active proposal stage.
Internal tool at a Costa Rica fintech · React + Google Apps Script
Serverless SPA that centralizes PTO requests, enforces blackout periods, and syncs approvals to Google Calendar. Replaced the manual email-chain process that came before it.
Personal project · Medallion architecture
End-to-end data warehouse following Bronze, Silver, and Gold layer separation. Cleaning, modeling, and analytics on top of SQL Server.
A small sample of the data side of the practice. Two Tableau Public projects, built to show what a non-technical stakeholder can diagnose on their own when the dashboard is designed for them, not for the analyst.
Tableau Public
Three coordinated views: workforce overview, demographics, and income analysis. Cross-filtered, drill-down from KPIs to employee records.
Tableau Public
Year-over-year sales and customer retention with seasonal trend isolation. Category and region filters, weekly profit averages.
I work between business operations and engineering. The early years were underwriting and data: finding where the numbers do not add up. The recent ones added engineering. When the off-the-shelf tools could not fix the problem in front of me, I started building the ones that could.
Today I supervise a 14-person team at a Costa Rica fintech, automating validation pipelines, dashboards, and reporting. The AI workflows I introduced are in daily use across the team. On the side, I design, build, and operate Nexo Barber, a multi-tenant SaaS for barbershops shipped end-to-end in Next.js and Supabase.
Underwriting taught me to assume hostile inputs and validate everything. Product taught me to ship the smallest version that still answers the question. I move fastest when those two instincts overlap: data work that has to be defensible, internal tools that change how a team operates, and product engineering for fintech.
Nexo Barber is shipped and in steady state. I keep side projects as a personal R&D lab where I test stack patterns and AI workflows, then carry the lessons back to my main work. I look for roles where side practice is a feature, not a conflict.
01
SQL-first work, from raw bank statements and Salesforce exports to the dashboards senior management actually opens. Window functions, validation pipelines, risk-pattern detection on top of it.
SQL · Python · pandas · Power BI · Tableau · Excel + VBA · Salesforce
02
Full-stack TypeScript on Next.js 16 and React 19, backed by PostgreSQL through Supabase. I default to multi-tenant from day one, with Row-Level Security, real-time channels, and PWA install paths. Mobile-first design systems, observability through Sentry, deploys on Vercel.
TypeScript · Next.js 16 · React 19 · PostgreSQL · Supabase · TailwindCSS · Framer Motion · Vercel · Sentry
03
When the off-the-shelf tools cannot fix the problem, I build the one that can. Internal apps in Google Apps Script and React. Macros for the recurring work. AI workflows orchestrated through Claude Code, custom skills, and MCP, for the parts of the job where the leverage actually compounds. The team I supervise has these in daily use.
Claude Code · Anthropic SDK · MCP · Custom skills · Memory systems · Google Apps Script · Excel + VBA · Python
04
Nine years underwriting merchant cash advances. Bank statement analysis, ACH and NSF pattern reading, fraud-indicator detection, pre-approval logic. The lens I carry into every product decision: assume a hostile actor on the other side of the form.
MCA underwriting · Bank statement analysis · Fraud detection · Pre-approval logic
Funding Metrics Sep 2021–Present
Funding Metrics Jun 2018–Sep 2021
Funding Metrics Jun 2017–Jun 2018
Universidad Hispanoamericana, Business Administration (2015–2016, Costa Rica).
Best for reaching me: email or LinkedIn. Replies within 1 to 2 business days.