Principal Engineer · Founder · 20+ years shipping production systems

I turn ambiguous problems into systems that hold up in production.

I build reliable agent systems at the boundary between probabilistic intelligence and deterministic software—from reproducible evaluation through production operation.

Jader Correa

Technical direction without leaving implementation behind

I am a principal engineer and founder. Across Brazil, North America, and Europe, I have remained hands-on while working as a staff engineer, technical leader, and former CTO.

My best work happens in small, strong teams where I can reduce uncertainty, shape architecture, raise the engineering bar, and own consequential systems through production.

Models can investigate, propose, implement, and act. Deterministic software retains authorization, product judgment, state transitions, and executable verification.

ENGINEERING AND LEADERSHIP EXPERIENCE
Me Salva!
Árvore
Code Climate
TransportAPI
Appmax
APOIA.se

Ideas expressed as working systems

Projects and production systems that show how I reason about reliability, product constraints, and engineering evidence.

Open source · Go · AI agents

GPTCode

An open-source, research-driven engineering project for explicit and verifiable coding-agent workflows. Its explicit workflows produce evidence, not just answers.

Explore GPTCode →
Evaluation research · Reproducible evidence

AI Experiments

An ongoing research program into coding-agent reliability, local-first inference routing, evidence-carrying handoffs, bounded session state, and semantic intermediate representations. Protocols retain failed and invalid runs and state exactly what the evidence supports.

Inspect the research →
Technical paper · Reliability

One Successful Agent Run Proves Almost Nothing

A repeatable protocol separating model capability from system reliability, with negative results, explicit limitations, and harness corrections.

Read the paper →
Engineering essay · Agent state

The Transcript Is Not the State

A systems thesis for separating observed evidence, working state, intent, effects, and language—so successor agents inherit inspectable state instead of an authoritative narrative.

Read the essay →
Founder · Production agent systems

Zapfy

An AI-native commerce platform operating with real merchants. Agents investigate and execute through a dedicated MCP service, while tenant isolation, authorization, idempotency, inventory, and financial and fiscal transitions remain deterministic.

Visit Zapfy →
Open source · Elixir · PostgreSQL

pgroll_ex

Elixir tooling that converts Ecto migrations into pgroll-compatible zero-downtime migration workflows.

View on GitHub →

Reduce uncertainty with executable systems

01

Start from the operational outcome

Architecture exists to produce a product and business result. I make constraints and trade-offs visible before scaling the solution.

02

Build reference implementations

A focused working system is often more useful than a large speculative design. It exposes assumptions, integration risks, and real costs.

03

Demand production evidence

Tests matter, but correctness also depends on observability, release behavior, stable contracts, and what the system does under real load.

04

Use AI without surrendering authority

Models can contribute reasoning and execution. Deterministic systems retain permissions, invariants, verification, and final authority.

Hands-on systems, measurable outcomes

Code Climate

Evolved the architecture of Velocity and shipped capabilities that contributed approximately $1M in annual recurring revenue.

TransportAPI / First Bus

Designed and built Elixir/OTP systems supporting real-time transportation applications used by more than 3 million passengers across 40+ UK cities.

AgroMais Cloud

Built an ML experiment involving industrial telemetry—exploratory work rather than a high-volume production deployment.

Árvore

Helped grow the engineering organization threefold while improving execution across products serving more than 2 million monthly users.

Principal engineering for consequential systems

I am interested in Principal Engineer and senior staff-level roles where I can remain hands-on, set technical direction, and own systems through production.

AI agents · agent evaluation and reliability · developer tools · distributed systems · PostgreSQL · Elixir · Go · TypeScript