Technical Lead · Senior Full-Stack Engineer · Applied AI
Agents, RAG, and multi-provider LLM architecture — engineered with the same rigor I bring to enterprise backends: safety, cost control, and clean abstractions that survive contact with production.
Based in Brazil · Open to fully remote roles worldwide
I'm an engineering team lead and senior full-stack engineer with 15+ years of professional software development — backend architecture, frontend engineering, DevOps, and cloud — shipping software for startups, product teams, and enterprise clients.
Today I lead engineers on complex client and internal projects while staying hands-on: I design and review architectures, review pull requests, mentor developers, and turn business requirements into architecture, milestones, and well-scoped tasks.
My technical focus is where software engineering meets AI: agent architectures, retrieval pipelines, and LLM integrations that are actually safe and affordable to run — plus AI-native engineering workflows where agents do real, reviewable work. I care about the unglamorous parts — context quality, PII boundaries, token budgets, evaluation, provider fallback — because that's what separates a demo from a system people can trust.
AI · RAG · Agents
I decompose initiatives into project → architecture → milestones → features → tasks → agent-executable work. Agents get the right context, constraints, tools, and acceptance criteria; they inspect the repo, run tests, and iterate on failures — with human review gates before anything ships.
A single interface that swaps between Anthropic, OpenAI, Azure, Gemini, and local models — with per-tenant encrypted keys, model selection per task, and automatic failover. The assistant on this site runs the same pattern live: a model roulette that fails over on the first token.
PII redaction at the prompt boundary, sensitive processing in an isolated agent that never touches the application tier, and fail-fast token-budget middleware that blocks over-limit requests before they hit the API.
Prompt and agent regression tests, tool-call and structured-output validation, retrieval-quality checks, and an immutable audit trail tracing provider, model, tokens, cost, and latency on every call — so failures are diagnosable, not mysterious.
Production LLM pipelines processing thousands of documents — chunking, embeddings, metadata filtering, structured JSON-schema output, batch APIs for major cost savings, and provider-agnostic orchestration with anti-misinformation gates.
Designing the information around the model: system instructions, repository knowledge, relevant-file selection, context compression, and token budgets. Plus Model Context Protocol servers, tool calling, and a published Claude Code skill.
How the AI layer is built
Swap Claude ↔ OpenAI ↔ local with no rewrite. Sensitive data is stripped before the prompt; over-budget requests fail fast; every call is logged with tokens, cost, and latency.
Selected Work
Civic-tech · data platform
An LLM pipeline that classifies news sources, extracts structured claims from thousands of articles, and evaluates public-record data — with cost-optimized batch processing and fact-check gating.
Construction · vertical SaaS
A vertical SaaS with an AI copilot — camera analysis, document intelligence, and streaming chat — built on a provider-abstraction layer with per-tenant budgets and PII redaction.
Data governance · SaaS
A cloud data-quality service with a dedicated processing agent isolated from the API, ensuring sensitive records are analyzed without ever crossing into the application layer.
Energy · enterprise BI
Enterprise business-intelligence dashboards and data integrations for a large energy utility — embedding analytics into operational tooling for non-technical stakeholders.
Fintech · payments
Role-based access control, permission-driven UI, and data-quality frameworks for a global payments organization — with security hardening and enterprise UAT.
Invoicing · SaaS
A multi-tenant electronic-invoicing platform integrating with tax-authority systems — certificate management, XML storage, quotas, and subscription handling.
Client and product names withheld by design.
Enterprise data & analytics consultancy
Current
Independent consultancy & own SaaS products
2021 – present
E-commerce, fintech, edtech, mobile
Earlier
Centro Universitário IESB
2015 – 2016
Universidade Bandeirante de São Paulo
2004 – 2007
Looking for someone who can take AI from prototype to production — safely, and at a sensible cost — and lead the team doing it? I'd love to hear about it.
Open to
Based in Brazil · Open to fully remote roles worldwide