AI Agents & Agentic Workflows
Systems that interpret inputs, choose bounded actions, use existing tools and request human approval when necessary.
Madeira, Portugal · Available globally
AI Automation Architect.
Production Engineer.
I design AI systems that automate real business workflows — safely, measurably and in production.
20+ years building software, distributed systems and production platforms.
What I do
AI that does work,
not just talks.
I build systems that interpret emails and documents, consult business knowledge, interact with existing software, execute workflows and leave an auditable record of what happened.
RAG & Enterprise Knowledge
Reliable access to internal documents and business knowledge with retrieval, citations, permissions and evaluation.
Business Process Automation
Workflows that move information, apply rules and complete repetitive operational steps without adding another dashboard.
AI Integrations
Connections between models, email, documents, APIs, CRMs and the software your organization already uses.
AI Architecture & Guardrails
Explicit scope, permissions, budgets, audit trails, fallbacks and controls for every autonomous action.
Backend & Cloud Systems
The APIs, queues, data flows, observability and deployment foundations that make AI useful in production.
The kind of problems I solve
Start with the work.
Then decide if AI belongs.
- 01Your team copies information between email, Excel and CRM.
- 02Customer requests require repetitive manual triage.
- 03Employees spend hours searching internal documents.
- 04Sales leads arrive but follow-up is inconsistent.
- 05Documents need extraction, classification or validation.
- 06Several tools exist but none of them communicate properly.
- 07A workflow requires human decisions but most steps are repetitive.
If a process happens repeatedly, follows rules and consumes human time, it is probably worth examining.
Not everything should be automated.
How I think about AI
AI in production is a
software engineering problem.
Building a demo is easy. Production introduces permissions, cost limits, failures, hallucinations, observability, retries, security, auditability, human approval, model changes and data quality.
AI is becoming easier to build. Reliable AI systems are not.
01
Observe
→02
Reason
→03
Act
→04
Verify
→05
Escalate
Every autonomous action should have:
ScopePermissionsBudgetLogsFallbackKill switch
Experience
20+ years building things
that have to work.
From backend software to technical leadership, regulated healthcare and AI systems — across SaaS, distributed systems, APIs, cloud and DevOps.
- 2005
Software Engineer
Built business platforms and learned production from the inside: users, failures, data and operational constraints.
- 2010
Technical Lead
Moved from implementation into architecture, delivery ownership and technical coordination.
- 2016
Engineering & Architecture
Designed APIs, integrations, CRM and workflow platforms for distributed teams and clients.
- 2022
Technical Manager
Led delivery and healthcare interoperability work in a regulated US HealthTech environment.
- 2024
AI Architecture
Defined an AI-enabled healthcare platform, coordinated engineering and AI specialists, and established cloud delivery practices.
- Now
Principal-level execution
Combine AI automation, backend architecture and hands-on production engineering from Madeira, Portugal.
Founder — Cre4tiva: independent product exploration alongside professional engineering work.
Selected work
Systems, not demos.
Most of my professional work is private. These summaries describe the problem and engineering decisions without exposing confidential code or customer data.
Professional
AI-enabled healthcare knowledge
- Problem
- Clinical information had to be made useful through an AI-assisted product without treating model output as unquestionable truth.
- Approach
- Defined technical requirements, coordinated engineering and AI work, and designed retrieval and LLM interaction flows.
- Architecture
- Azure, Python, LLM workflows, vector search, Docker and CI/CD.
- Outcome
- Delivered a secure platform foundation from concept through deployment, with bounded AI workflows for medical information use cases.
Professional
Healthcare interoperability
- Problem
- A regulated platform needed dependable connections with multiple healthcare providers and interoperability systems.
- Approach
- Coordinated delivery, vendors and stakeholders while implementing FHIR, HL7 and third-party API integrations.
- Architecture
- AWS, FHIR, HL7, REST APIs, SQL and production observability.
- Outcome
- Improved the reliability and visibility of critical integration workflows while supporting HIPAA and HITRUST requirements.
Professional
Legacy platform modernization
- Problem
- A mature municipal operations platform needed new delivery without destabilizing long-running production behavior.
- Approach
- Modernized backend code, maintained APIs and jobs, diagnosed regressions, and strengthened automated tests and deployment safety.
- Architecture
- PHP, MySQL, REST APIs, Docker, background processing and CI/CD.
- Outcome
- Supported safer ongoing delivery while progressively modernizing production-critical components.
My principles
Useful constraints.
01Not every problem needs AI.
02Agents should have less authority than the humans supervising them.
03Automation should remove work, not create another dashboard.
04If you cannot observe an AI system, you cannot trust it.
05AI should fail safely.
06A good AI architecture assumes the model will eventually be wrong.
Writing / thinking
What I'm thinking about.
AI automationAI agentsAI safetyProduction engineeringArchitectureVibe codingEngineering leadership
Published book · Available on Amazon
La Mentira del Vibe Coding
La IA puede escribir tu código. Tú tendrás que responder cuando falle.
A practical argument for keeping engineering judgment, architecture and accountability at the center of AI-assisted software development.
View the book on Amazon →Work with me
OPTION A
Need AI automation?
Tell me about the process your team keeps doing manually.
Automate a Process →OPTION B
Hiring?
I'm open to Principal, Staff, AI Architect, Forward Deployed and senior engineering roles where architecture and execution matter.
Hire Me →Prefer a direct professional channel? Connect on LinkedIn or download my CV.