Senior engineer with 15 years of experience in technology. I take fixed-scope, async projects and deliver them as code or a written report:

• Cloud and DevOps: AWS, Google Cloud, Terraform, Kubernetes and CI/CD pipelines with GitHub Actions.
• AI with a person in control: n8n automations with a human approval step, PDF chatbots that cite their sources, and review and fix of apps built with AI coding tools.

AWS Certified Solutions Architect – Professional · Google Cloud Professional Cloud Architect · Certified Kubernetes Administrator (CKA)

I built these to show how I work. The companies in them are fictional and none of it is client code. Each one runs on your own machine with no cloud account, and its README lists what was not tested.

Cloud, Terraform and Kubernetes

• github-actions-safe-deploys: GitHub Actions that build an image once, scan it, wait for an approval, promote it by digest and roll back, tested on a throwaway kind cluster. The AWS and Google Cloud login steps never ran against a real account.
• gcp-cloud-run-keyless-deploy: the PDF chatbot packaged for Google Cloud Run in Terraform, set up to deploy through Workload Identity Federation instead of stored keys. Tested offline; never applied to a real project.
• kubernetes-helm-release-tested: a Helm release installed, upgraded, rolled back and checked for drift on a throwaway cluster in GitHub Actions.
• terraform-aws-review-before-after: a deliberately flawed AWS Terraform setup, each risk fixed in its own commit with a policy check that fails before and passes after.
• nextjs-aws-terraform-signed-video: private lesson videos behind CloudFront signed links; AWS infrastructure in Terraform, checked by terraform test with a mocked provider in GitHub Actions. Not deployed to a real account.
• aws-delivery-gap-assessment: method, read-only evidence kit and sample report for checking contracted AWS work against what actually runs.
• k8s-agent-readonly-lab: what actually stops an AI agent from changing a Kubernetes cluster, tested on a throwaway kind cluster.

AI automation and AI-built apps

• ai-built-app-rescue: a demo booking app written with typical AI-code mistakes, then fixed. One fix per commit, each with a test that fails before and passes after.
• supabase-rls-tenant-audit: scripted tests that show whether tenant A can read tenant B's data.
• n8n-sales-followup-agent: AI drafts, a person approves, nothing is sent twice.
• pdf-rag-citations: a PDF chatbot that cites the page and refuses to guess.
• reliable-ai-jobs-api: background AI jobs that retry safely, survive a crashed worker and check a spending cap before every call.

Fixed price, async. On Upwork: Tales Andre Costa V. · AI-built app review and fix