Principal Cloud Architect · AI Systems Engineer
I build what others can't.
I fix what they broke.
For 14 years, I've engineered mission-critical systems that outperform multi-million-dollar corporate vendor software. Now, I architect scalable AWS infrastructure and deterministic multi-agent AI workflows for enterprises that need things to actually work.
The Baseline
From Grid-Scale Infrastructure to Deterministic AI.
8 years as a Senior Electrical Engineer at ONS — Brazil's grid operator, powering 200+ million people. I don't tolerate inefficient architecture. I delivered a serverless load management platform in 4 months that still outperforms its R$ 6M corporate replacement. I built an async API that IT adopted to replace the vendor's standard tooling. I took a 15,000-line Fortran monolith and refactored it into the official modular open-source model. If it doesn't scale, I rewrite it.
In 2022, I started architecting AgendaCheia — a massive SaaS marketplace. Working 03:10 AM deep-work shifts, the codebase scaled to a staggering 14 million lines of TypeScript. At that scale, every bug is a critical failure. The cognitive load exceeded human limits. The standard AI tools were useless toys. I realized a fundamental truth: AI hallucination is an architecture and context problem, not a model problem. I designed SpecForge — an MCP-native AI orchestrator running 50 parallel agents with 5-level review gates. Today, I don't just write code; I orchestrate deterministic software factories.
Systems Built
Deployed Architecture.
I don't build disposable MVPs. Every system here was engineered to solve a critical scale or automation bottleneck.
Solutions Forge
Founder · 2026A high-end software engineering firm shipping production platforms, intelligent automation, and scalable cloud systems. We architect it, we build it, we maintain the infrastructure.
solutionsforge.tech →SpecForge
Beta · 2025Enterprise-grade AI agent orchestration for software engineering. MCP-native architecture, 57 integrated tools, and wave execution using 50 parallel agents with deterministic, multi-level review gates. The engine behind a 14M LoC codebase.
AgendaCheia
Beta · 2022A 14M+ LoC TypeScript scheduling SaaS platform. 8 apps, 17 shared packages, Stripe Connect marketplace. Built entirely by one engineer dogfooding the SpecForge AI orchestration engine.
PipeForge
In DevelopmentAWS-native Data and ML pipeline manager. Orchestrating ETL, training, and deployment on AWS Batch and Fargate to bypass massive SageMaker cold-start costs.
ONS — National Grid Systems
2017 — PresentMission-critical ML and data engineering pipelines serving the operational backbone of Brazil's continental-scale power grid. Built entirely outside of my job description to replace failing legacy software.
Stack
The Engineering Arsenal.
No religion. Just the most efficient tools to achieve deterministic scale.
Languages
- TypeScript
- Python
- R
- C#
- SQL
Frontend
- Next.js
- React
- Tailwind CSS
- Shadcn/UI
Backend
- Node.js
- Stripe API
- MCP Protocol
- REST / GraphQL
Cloud Architecture
- AWS (Lambda, Batch, ECS, S3)
- DynamoDB
- Amplify
- Cloudflare
AI Orchestration
- Claude API
- Agent Orchestration
- Forecasting Systems
- Statistical Modeling
Infrastructure & CI/CD
- Claude Code CLI
- SpecForge MCP
- Git / GitHub
- Turborepo
Telemetry
System History.
Initialization
First lines written in Fortran and VBA. Started automating engineering calculations because manual work is a failure of imagination.
Grid-Scale Operations
Joined ONS. Immediately recognized the inefficiency in multi-million-dollar vendor tools and began architecting custom data pipelines that actually scaled.
The Cloud Transition
Refactored legacy monoliths into AWS serverless architecture. Mastered the TypeScript ecosystem to begin the architecture of AgendaCheia.
The Crucible
Gramado Summit launch aborted due to a family medical emergency. Four months offline. The capital was burned, but the architectural blueprints remained intact. Rebuilt momentum with zero tolerance for wasted time.
The AI Breakthrough (SpecForge)
AgendaCheia scaled to millions of lines of code. The human bottleneck was reached. Diagnosed standard LLMs as fundamentally flawed for software engineering due to context decay. Architected SpecForge: an MCP-native, 50-agent parallel orchestrator. 400+ builds later, the 14M LoC problem is solved.
Execution (Solutions Forge)
The countdown to July 2026 begins. The infrastructure is set.
He didn't know it was impossible, so he did it.
Advisory
Initiate Protocol.
Building something that needs to survive contact with reality? Have an architecture bottleneck and no patience for consultants who've never shipped 10M+ LoC? Let's talk.
"There is no secret.
Just a choice you make
at 03:10 every morning."