ferrolho.
EN/PT
AI ENGINEERING · PLATFORMS · REAL-WORLD DELIVERYLisbon, Portugal

Hello, I’m Rui.

AI ideas.
Engineered into reality.

I turn ambitious ideas into working platforms. From secure app deployment to knowledge systems and AI agents that ship code — with people in control.

Explore my projectsLet’s talk ↗
RUI FERROLHO↓AI ENGINEERING / V.02
01SELECTED PROJECTS

Less theory.
More working software.

Three projects I founded and built at Enhesa, from the first idea to real use or a working proof of concept.

01 / FORGEINTERNAL PLATFORM
01An idea
02A short brief
03Forge
04Test + Production
Dozens of appsOne platform. The infrastructure taken care of.
AI-ASSISTED PLATFORM ENGINEERING

Forge.

From an idea to a secure, live app.

AI made writing an app easier. Getting it safely into production was still taking weeks. I founded Forge and built the proof of concept solo, with Claude Code as my pair programmer.

Now a team-run platform hosting dozens of internal apps at Enhesa.

Inside the project

A short form generates a prompt for a coding agent. Forge creates the Git repository and build pipeline, deploys to Kubernetes, and separates Test from Production.

Company single sign-on, outbound network restrictions, daily code and data backups, and monitoring are built into the platform.

The proof of concept secured company investment. Today, teams build internal apps without having to learn Kubernetes.

What I learned

As AI writes more of the code, the engineering value moves to the guardrails: security, cost and reliability.

BUILT WITHTypeScript · Next.js · PostgreSQL · Redis · Kubernetes (AKS) · Helm

02 / LUMENDEPLOYED INTERNALLY
01Pages
02Hybrid search
03Relevant passage
04People + Agents
3½ weeksMeaning + keywords. Permissions at every step.
KNOWLEDGE SYSTEMS FOR AI

Lumen.

The right context. At the right moment.

AI assistants need knowledge that is often scattered across notes and people’s heads. I founded Lumen and built it as its sole developer: a knowledge base designed around what agents need to find.

Deployed internally at Enhesa after around three and a half weeks of development.

Inside the project

Each saved page is split into passages and embedded. Hybrid search combines meaning with keywords and returns the best passage, keeping agent context small.

Permission-aware results power the search bar, a chat with source citations, and Claude Code through MCP. AI-written index pages give agents a map of each folder.

Live co-editing, revision history with rollback and pull-request-style approvals support the people behind the knowledge.

I wrote the brief and technical design, then directed Claude Code one milestone at a time, tests before code: 13 milestones, over 1,000 commits and more than 10,000 automated tests.

What I learned

An agent is only as good as the context it can find. Storing knowledge is easy; making it findable is the real work.

BUILT WITHTypeScript · Node.js · React · TipTap · Yjs · PostgreSQL / pgvector · Claude · MCP · Docker

03 / FACTORYOSPROOF OF CONCEPT
01Request / Jira
02Plan · Code · Test
03Human approval
04Draft pull request
Linux + WindowsReal builds. Review evidence. Human approval.
AGENTIC SOFTWARE DELIVERY

FactoryOS.

Agents do the work. People hold the keys.

Even a small code change means understanding a repository, getting it building, testing and preparing a review. I founded and built FactoryOS solo to hand that loop to Claude agents, with human approval built in.

Running end to end on Azure. A team of six is adopting the proof of concept.

Inside the project

Agents analyse repositories, plan changes, write code and run real builds and tests on Linux and Windows, including legacy .NET Framework applications.

Questions go to the team’s knowledge base first, then to the engineer. Separate code and security review agents, plus a secret scan, check every change.

Nothing is pushed until a human approves. The resulting draft pull request includes test logs, review findings and cost. Project isolation, spend caps, a Stop button and a kill switch bound execution.

The platform has built and tested a real legacy .NET application on Azure. Adoption by a team of six is now underway as a proof of concept.

What I learned

Safety belongs in code, not prompts. Agents propose; fixed rules and a human decide what ships. Evidence beats claims.

BUILT WITHTypeScript · Node.js · Next.js · Fastify · PostgreSQL · Temporal · Docker · Azure · Claude Agent SDK · GitHub · Jira

OPEN SOURCE

Big Agentic Brain ↗

I also help maintain Big Agentic Brain: shared memory and coordination for AI agents.

Explore on GitHub ↗
02HOW I CAN HELP

Let’s solve the right problem.

01

From idea to proof

Turn a real problem into a working prototype, test assumptions and define what comes next.

02

AI with context

Connect agents to the right knowledge through search, permissions and useful workflows.

03

From demo to delivery

Build the foundations for running applications: authentication, deployment, monitoring and human control.

Tell me about your idea ↗
03THE PERSON BEHIND THE PROJECTS
Rui FerrolhoRUI FERROLHO / LISBON

I’m Rui.
I like making ideas work.

I work at the intersection of software engineering, platforms and AI. At Enhesa, I founded Forge, Lumen and FactoryOS and built their initial versions solo, using AI as a development partner.

The thread through my work is simple: make technology useful while keeping security, reliability and human control in view. I write the technical design, direct the implementation and look for evidence that it works on real systems.

My background includes studies at Politécnico de Leiria (2016–2018), with a final project in Information Systems, and training in project management.

Open to new projects, collaborations and opportunities.
04LET’S BUILD SOMETHING

Your next idea.
Let’s make it work.

Have a project in mind, a problem worth solving, or an opportunity to explore? I’d like to hear about it.

Say hello on LinkedIn
EMAILrui@rsolutions.ptWHATSAPP+351 965 476 369CALL ME+351 965 476 369