AI Agent Architect

João Moura Portfolio

Founder of CrewAI

Founder of CrewAI, an open-source framework (59,000+ GitHub stars) for orchestrating role-based teams of autonomous agents that collaborate on a shared task, with each agent assigned a role, goal, and backstory; previously worked in backend engineering at Clearbit before moving into agent framework design.

Multi-Agent SystemsAgent OrchestrationCrewAIOpen Source

Location

Not listed — probably too busy shipping to update the bio.

Companies

CrewAI

Skills

Multi-Agent SystemsAgent OrchestrationCrewAIOpen Source

Notable

  • CrewAI: 59,000+ GitHub stars

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

A role-based abstraction for multi-agent coordination

Structuring agents around explicit roles, goals, and backstories, rather than an undifferentiated agent pool, gives multi-agent systems a mental model borrowed from how human teams are organized, part of why the framework was adopted quickly.

2

Adoption scale as an honest, third-party signal

Star count (59k+) is a visible, unmanipulated measure of how many developers actually chose this orchestration approach over alternatives, not a self-reported claim of impact.

3

A visible path from general backend engineering into a specialized architecture niche

A background in general backend systems (Clearbit/HubSpot) rather than an ML research pedigree shows agent orchestration is also a legitimate specialization path for systems-minded backend engineers, not only ML researchers.

What AI Agent Architects can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • Give your multi-agent framework a clear organizing metaphor (roles, teams, goals) — it makes the architecture easier for other developers to reason about and adopt.

  • Let adoption numbers (stars, downloads) speak for themselves on your profile as one honest signal, rather than only describing impact in your own words.

  • If you're moving into agent architecture from general backend or systems engineering, say so — it shows the specialization is a deliberate path, not a requirement to have started in ML research.

Projects

Years of small, focused libraries came before the agent framework: an Elixir state machine and a Rails gamification gem. CrewAI's role-and-task model comes from the same habit of giving a messy problem a small, declarative shape.

Open SourceadvancedOngoing (active open-source project)

CrewAI

PythonMulti-agent orchestrationEvent-driven workflowsLLM providersMCP

A Python framework for multi-agent automation with two layers. Crews are teams of role-based agents, each with a role, goal, backstory and tools, working together autonomously. Flows are event-driven workflows for precise control that can call Crews as steps.

The role, goal and backstory abstraction lets non-specialists understand multi-agent design. Adding Flows answered the main production complaint about autonomous crews: they are unpredictable.

  • Crews for autonomy, Flows for deterministic control, and the two compose

  • Agents declared by role, goal, backstory, tools and memory

  • README: 100,000+ developers certified via learn.crewai.com

Open SourceintermediateOpen source; pinned on his GitHub

Machinery

ElixirPhoenixEctoHex

A lightweight state-machine library for Elixir with built-in Phoenix integration. A small DSL declares the states on a struct, and guard functions, before/after callbacks, persistence hooks and transition logging control how state changes.

Explicit states, guarded transitions and logging are the same concerns that make agent workflows debuggable. He was building them into libraries well before agents.

  • Declarative states on plain Elixir structs

  • Guard functions plus before/after callbacks per transition

  • Phoenix integration and a state dashboard

Open SourcebeginnerOpen source; 2.0.0 RC

Gioco

RubyRuby on RailsActive RecordRubyGems

A Ruby gem that adds gamification to Rails apps using Active Record, with badges, levels and points or currencies as ready-made models. You install it from RubyGems with one Gemfile line.

An early example of his pattern: pick a domain concept, turn it into a small reusable abstraction, and publish it for others.

  • Badges, levels and points as drop-in Active Record models

  • One-line Gemfile install

  • Versioned API with the legacy 1.x kept on its own branch

Ready to build your portfolio?

Follow the AI Agent Architect roadmap — skills, projects, and timeline to get hired.

AI Agent Architect Roadmap