$ exec --since 2019

EXEC Labs builds production AI systems for high-growth companies and develops its own portfolio of products. Headquartered in São Paulo, working with teams across the US and Brazil.

  • 2019building software since
  • 2024AI agents in production since
  • US + BRclients across both markets
  • 6industries served
scroll

// what EXEC executes

Three lines of business.
One engineering core.

Production AI for high-growth companies, our own portfolio of products, and selective partnerships. All of it runs on the same core: senior engineering that ships and measures.

01 · Engineeringactive

Production AI for high‑growth companies

agents · MCP · evals · cost

LLM agents and workflows, MCP servers and tool gateways, evals and observability, LLM cost engineering, and full-stack product engineering (TypeScript, React, Node, cloud and edge).

02 · Products3 on the network

Our own portfolio

mobile · desktop · web

EXEC Labs designs, builds and publishes its own mobile, desktop and web products. They are where our engineering gets tested first.

03 · Partnershipsselective

Selective partnerships

co-founding · builds · audits

Technical co-founding, focused builds, architecture and AI audits.

// selected work · telemetry

Selected work.

Two industries, two markets, both under NDA. The numbers come from production.

2019Building software since 2019
2024AI agents in production since 2024
US + BRClients across both markets
6Industries: agribusiness, energy, fintech, healthcare, e‑commerce, legal

// fintech · United States · since 2025

Engineering partner to a venture-backed US AI company automating mortgage-loan processing.

Under NDA
Agent platform 0+ production AI workflows
MCP gateway ~0 tools behind an OAuth 2.1 MCP gateway
Tool-call failures 44% → <10% tool-call failure rate
LLM cost −80–85% LLM cost on the highest-spend workflow
  1. Built the AI agent platform from scratch: 13+ reactive and proactive workflows on five event queues, with confidence gates that act autonomously on high-confidence work and route the rest to people for approval, under full audit logging.
  2. Designed the MCP gateway that gives AI agents (Claude, GPT, Cursor) secure access to ~90 data tools; standardized schemas cut tool-call failures from 44% to under 10%, and response shaping cut tool payloads by ~80%.
  3. Engineered layered LLM cost controls (deterministic pre-filters, risk fingerprinting, multi-provider model routing) and cross-provider fallback across Anthropic, OpenAI and Google.
  4. Built the quality layer: per-workflow tracing with cost breakdown, eval suites (deterministic and LLM-as-judge) and detection of silently missed runs.
  5. Shipped an event-driven audit-log microservice in one week, with field-level AES-256-GCM encryption of personal data.
  6. Rebuilt search end to end: from 3–10 s to 50–100 ms.

// agribusiness · Brazil · 2024–2025

Technical leadership of an AI platform for agribusiness, from the agent framework to the product.

Under NDA
Delivery speed +0% faster feature delivery
Code duplication −30% code duplication
Retention +0% user retention
Model providers 3 model providers behind one agent framework
  1. Built a provider-agnostic AI agent framework (OpenAI, Anthropic, Gemini) with specialized vertical agents and seamless handoffs, before off-the-shelf TypeScript agent frameworks had matured.
  2. Designed a generative-AI conversational BI tool that renders live data visualizations inside the chat, so farmers query field operations in natural language.
  3. Led the full-stack platform in a TypeScript monorepo (React, Next.js, Fastify, tRPC, hexagonal architecture).
  4. Shipped passwordless OTP authentication, lifting user retention by 25%.
  5. Delivered alongside a distributed team across five countries.

// industries we've served

  • Agribusiness
  • Energy
  • Fintech & mortgage
  • Healthcare
  • E-commerce
  • Legal

// operating procedure

How we work.

No account managers, no handoffs, no juniors learning on your project.

Senior-led

The people in your calls are the people in your codebase. Decisions don't get lost between layers.

Async-first

US-hours overlap, written decisions (ADRs) and weekly demos, from UTC−3.

Measured

We ship to production and measure it: evals, traces and costs, so "it works" comes with numbers.

// portfolio on the network

Products by EXEC Labs.

Beyond building for others, EXEC is also home to its own brands and products. That was true in 2020, and it's the plan now.

Beta

termx

SSH / mosh terminal built as a mobile control surface for AI coding agents.

Android
Live

Artifact Hub

Share AI-generated files at links that render in the browser.

Coming soon

MyPocketGym

Evidence-based training, with grounded AI coaching.

AndroidWeb
R&D

More in the lab

New products in active development.

mobiledesktopweb

// company · flight log

From hackathon to headquarters.

Building software since 2019: from early innovation awards to a headquarters on Avenida Paulista, serving clients in the US and Brazil.

  1. Phase 12019Origin

    Recognized early.

    Awarded and selected in our very first year.

    • 1st place in an innovation hackathon

      With a business-intelligence solution for an energy cooperative, and a place in a technology incubator.

    • EXEC is born

      In Rio Grande do Sul.

    • Three EXEC projects selected by Programa Centelha RS

      Among them, Delivchat and e-Rural, in the innovation-entrepreneurship program run by Fapergs in Rio Grande do Sul.

  2. Phase 22020–2023Product engineering

    Platforms that move numbers.

    Web and mobile platforms for clients in e-commerce, healthcare, legal and field research, plus our first products: Delivchat, e-Rural and BI Systems.

    // outcomes
    +35% conversion and +20% average order value Mobile commerce app (React Native) versus the client's mobile site.
    −90% data-entry errors Offline-first field-research app; analysis time from weeks to hours.
    +40–50% form completion Lead capture for a law firm.

    Also: business-intelligence dashboards, a health-plan portal with a lead-qualifying chatbot, and an over-the-air mobile release pipeline that ships fixes without app-store delays.

    “Além de desenvolver soluções para terceiros, a Exec também engloba outras marcas e negócios.”

    Beyond building for others, EXEC is also home to its own brands and products.

    EXEC website, 2020
  3. Phase 32024–todayAI in production

    From agents to the frontier.

    Production AI for high-growth companies, and our own products.

    • 2024

      AI agents in production

      Leading an AI platform for agribusiness.

    • 2024

      Headquarters in São Paulo Av. Paulista

      On Avenida Paulista (EXEC Vanguard Solutions Ltda).

    • 2025

      US expansion

      Engineering partner to a venture-backed US AI company in fintech.

    • 2027

      Products

      EXEC Labs starts publishing its own portfolio.

// four letters, four principles

E · X · E · C

E

Explore

Work at the frontier, where the tools are new and the answers aren't written yet.

X

eXperiment

Prototypes and our own products: we test ideas in-house first.

E

Engineer

Take it to production, with evals, traces and costs in view.

C

Compound

Every product and every engagement adds up to the next one.

// open a channel

Something worth executing?

Production AI, a product partnership or a technical audit: tell us what you're building.

↵ click the line to copy the address

Start a project

Headquarters: Av. Paulista, 1471, São Paulo, Brazil · UTC−3