Skip to content
Portrait of Erick Gutiérrez, full stack developer and QA analyst

Hi, I'm

Erick Gutiérrez.

Full Stack Developer | Functional & QA Analyst

Scalable solutions.

Colombia · Remote

Years in financial QA
+2
Year working with AI
+1
Projects running in production
2
Right now

I develop and maintain web applications in the healthcare sector on technologies such as PHP with CodeIgniter 3. Alongside that I work independently as a full stack developer with React, Node.js, Tailwind CSS, Vercel, Supabase and PostgreSQL. Among my projects is a chatbot built for Unisangil's Nursing Faculty.

See projects
How I work

I use Claude constantly in my working environment, taking advantage of its whole ecosystem: Cowork, Claude Code, Artifacts, plugins, skills, subagents and agents. I always secure and review the code and everything the AI produces before it reaches production. The AI programs, I develop: it's a tool that optimises and maximises the speed and quality of the work, but I'm the one in control.

How I build it
Quality

I have more than two years in quality assurance in the financial sector, validating and maintaining the quality of transactional flows for Grupo Aval from the Postilion transactional switch, with ISO 8583 and Base24 messaging, ensuring transaction traceability. From the BUS-CORE integration layer, testing and validating microservices and their integrity against MySQL and Oracle SQL databases.

Where it's been tested
Availability

I work under agile methodologies, with Scrum certifications, and I'm always open to new opportunities where I can apply my combined development and quality experience. I'm currently finishing my Systems Engineering degree and I have conversational English for international opportunities.

Let's talk

Projects

What I build

A sample of recent builds where product delivery and technical rigor work together.

Case studies

Production

Nursing ChatBot — Lab Booking SaaS

Problem
The School of Nursing at Unisangil booked labs over email, WhatsApp and spreadsheets: a manual process, with no clear availability and prone to scheduling clashes.
Solution
A PERN platform that centralizes everything: a conversational assistant walks the teacher through creating a booking, a calendar shows real-time availability per area, and an admin panel manages users and bookings.
Technical decisions
  • Transactional concurrency controlthe booking is validated and confirmed inside a transaction to prevent double bookings when two teachers race for the same slot.
  • Role matrix (superadmin / coordinator / teacher)explicit permissions per action; a teacher creates their bookings but can't cancel them —that's the coordinator's responsibility— mirroring the school's real process.
  • Closed model, no public sign-upevery account is created by the superadmin, fitting an institution.
  • Real-time calendar (Supabase Realtime)availability updates live across users without reloading.
Role / Outcome
Solo project, end to end. Shipped and in production with the school; live demo.
React 19Vite 7Tailwind v4Node.jsExpress 5PostgreSQLSupabase RealtimeJWT
Production

FlowCheck — AI-Powered Test Automation

Problem
E2E testing tools (Testim, Mabl) are expensive and force you to maintain brittle selectors that break with every layout change. I wanted to show that a flow described in natural language can be validated in a real browser, without writing test code.
Solution
The user picks a flow type (login, signup, search, navigation or form), pastes the URL and the data. Gemini translates that into a step plan; a worker runs the steps in real Chromium with Playwright, captures a screenshot per step and streams progress live to the dashboard.
Technical decisions
  • Behavior-based verification, not selectorsinstead of trusting the AI's literal selectors, it resolves fields tolerant to language/markup and confirms by observed outcome, so tests don't break when the HTML changes.
  • Decoupled Vercel + Render architecturethe serverless frontend/API (seconds-long limit) triggers a long-running Express worker for 30–60s jobs, sidestepping the serverless timeout.
  • 3-layer SSRF protectionsince it navigates user-supplied URLs, it validates and blocks internal/private targets before opening the browser, preventing the tool from attacking the internal network.
  • Realtime + refetch reconciliationit combines Supabase Realtime events with a backup refetch, so live progress isn't lost if a WebSocket event drops.
Role / Outcome
Solo project end to end: design, AI, worker, infrastructure and tests against real Chromium. In production, with a demo and public repository.
Next.js 16TypeScriptTailwind v4SupabaseGemini APIPlaywrightExpressVercelRender

More projects

  • Multi-tenant SaaS for delivery companies: live maps with Mapbox, roles and subscriptions. Team project; superadmin complete.

  • Personal finance for Colombia: web (Next.js) + mobile (Expo) on one Supabase with RLS and shared Zod validation.

  • My first project, during my training at SENA: the foundation of my current stack.

The AI codes, I develop.

Coding is writing code. Developing is understanding a problem and delivering a real solution.

I direct the AI; I don't depend on it. The stack, database logic, architecture and repository are my calls. The AI speeds up execution.

Division of labour

I decide

  • Database logic and data model
  • Stack and project structure
  • Architecture and technical decisions
  • Repository and branch management

The AI accelerates

  • Spec-driven implementation
  • Boilerplate and repetitive refactors
  • Documentation
  • Exploring options and quick questions

Before any code is written

The AI can't decide what needs to be built. That work comes first, and it's mine.

  1. Understand the problem

    What hurts today, for whom and why. Without this, any solution is just a well-written guess.

  2. Pull out the requirements

    Translate what's asked into what's actually needed: business rules, the data involved and the constraints that can't be broken.

  3. Write the spec

    A short document fixing the scope, the technical decisions and the reason behind each one. It's what the AI executes: a bad spec produces bad code.

  4. Acceptance criteria

    What has to happen for the task to be done, and what must never happen. This is where the QA eye comes in: edge cases are defined up front, not afterwards.

  5. Document the decision

    The spec and the plan are versioned next to the code. Picking the context back up weeks later doesn't depend on my memory or a chat history.

The ecosystem, not just Code

Design

Claude Design

Mockups and visual exploration before writing code: layout, hierarchy and color.

Context

Claude Chat

A Project holds the app's context: I resolve questions and weigh decisions without re-explaining everything.

Documentation

Claude Cowork

Product documentation: specs, manuals and supporting material that travels with the code.

Implementation

Claude Code

Executes the plan: agents for scoped tasks and skills that lock in the repository's conventions.

Discipline and control

Model matched to complexity
Sonnet for the scoped and mechanical: boilerplate, refactors, changes where the judgment call is already made. Opus when the task demands judgment: architecture, specs, or a bug that won't reproduce. Spending the expensive model on something trivial costs as much as using the cheap one on a structural decision.
Effort level
I raise the reasoning when a decision is expensive to reverse and lower it when the result can be checked at a glance. Effort follows risk, not prompt size.
Context hygiene
I close and reopen the conversation when switching tasks. Carried-over context degrades answers and makes every message costlier; the spec lets me resume without dragging noise along.
A living CLAUDE.md
The repository's conventions live in the repository and are updated with it. That way I don't repeat the same correction every session or rely on the AI guessing the project's style.
Specs and plans kept apart
Each does one thing: the spec decides, the plan executes. Concise and separate, they serve to resume the work, to review it, and to let someone else understand why it's built this way.

That's the method. What follows is where it's been put to the test: payment systems in production, regulated certification, and the training behind it.

Journey

Where it's been tested

The method above doesn't live in a vacuum: today on healthcare applications, and before that certifying systems that move real money under Banco de la República regulation.

Current role

Full Stack Developer · Healthcare

Go Ecosystem · Colombia · Remote

Present

Jul 2026 — Present

Development and maintenance of web applications for the healthcare sector, on a legacy PHP base with CodeIgniter 3.

  • Healthcare web applications

    New features and evolutionary maintenance on PHP / CodeIgniter 3, with the same quality judgment I bring from the financial sector: edge cases are defined before writing code, not after.

    PHPCodeIgniter 3

Previous role

QA Analyst · Financial Sector

SQA S.A. · Colombia · Remote

Apr 2024 — Jul 2026

Certifying mission-critical payment systems across the Grupo Aval ecosystem, under regulated certification environments with the Banco de la República.

  • SPI · Instant Payments System

    I certified all 5 Grupo Aval entities (DALE, BOCC, BPOP, BAVV, BBOG) to operate on Colombia's SPI, validating P2P, P2M and QR flows over ISO 8583 traces (MTI 0200/0210) on Postilion and Base24.

    ISO 8583PostilionBase24MTI 0200/0210
  • PSE Avanza · ACH Colombia

    I validated the BUS↔CORE integration under the ACH Colombia standard with 180+ cases and 100% coverage of critical flows, reconciling messaging against Oracle.

    ACH ColombiaOracleSoapUI
  • Automation & Performance

    Automation with Serenity BDD + Cucumber (Screenplay pattern), performance testing with JMeter and WAPT, and Continuous Testing on Azure DevOps.

    Serenity BDDCucumberJMeterWAPTAzure DevOps
Entities certified
5/5
Cases per cycle
250+
First-pass approval
98%
Critical defects in prod
0

Education

  • In progress

    Systems Engineering

    University Education

  • Certified

    Scrum Certifications (SMC, SFPC)

    Scrum Certifications

  • Active participation

    Research Seedbeds

    QA / Performance / Automation

  • Completed

    ADSO Technologist

    SENA

What I build it with

QA & Quality

  • ISO 8583
  • Postilion
  • Base24
  • Serenity BDD
  • Cucumber
  • JMeter
  • Trazas Transaccionales
  • Cámaras de Compensación

Postman · SoapUI · WAPT · Playwright · Oracle SQL · Java

Regulated certification with the Banco de la República

Development

  • React
  • Node.js
  • PHP
  • CodeIgniter 3
  • Tailwind CSS
  • Supabase
  • PostgreSQL

NextJS · Vue · ExpressJS · MySQL · JavaScript · CommonJS

Two personal projects running in production

Delivery & AI

  • Claude Code
  • Azure DevOps
  • CI/CD
  • Git
  • GitHub

Antigravity · Vercel · Render

The workflow described in the section above

Contact

Let's build something solid.

Tell me about the project - I usually reply within a day.

Or reach me directly

LocationColombia · Remote

Availability

Open to new projects

As a full stack developer or as a Senior QA.