From a WhatsApp Group to an AI Think Tank

How six computer science friends from the late '90s turned a happy hour chat into a collective exploration of artificial intelligence — and why we decided to share what we learn along the way.

It started, as many good things do, with a simple question: "Can everyone make it on December 14th?"

In late 2023, three of us created a WhatsApp group to organize a night out. We'd studied Computer Science together at PUC-SP in São Paulo back in the late '90s — the kind of friendships forged over late-night coding sessions, heated debates about operating systems, and a shared conviction that technology could change the world. We hadn't lost touch, but life had scattered us across different industries and, in one case, across continents.

The plan was VR gaming in Moema, followed by drinks at a bar nearby. Classic reunion stuff.

But somewhere between booking the VR session and arguing about who would drive, the conversation drifted — as it always does with us — into technology. Someone shared a link about a new language model. Someone else asked whether you could run it locally. A third person was already spinning up a Docker container to find out.

The happy hour group became a daily feed of links, experiments, and arguments. Within weeks, we weren't organizing drinks anymore. We were building things.

The Cast

We're six people with very different day jobs but a common obsession.

André is an entrepreneur who, in his spare hours, runs self-hosted AI infrastructure and builds ML models for financial forecasting. He's the one who registered the domain, set up the servers, and keeps reminding everyone that models need a leash, not autonomy. Ask him what he does and he'll tell you he's "just the janitor" — but if there's duct tape holding this operation together, he probably applied it at 2 AM.

Guilherme is a senior software engineer and a university professor, but his real passion is squeezing every last drop of value from AI coding tools. He's the group's scout — constantly discovering new tools, frameworks, and workflows, then stress-testing them before anyone else gets a chance. If something useful exists in the AI tooling ecosystem, Guilherme has probably already tried it and filed a bug report.

Gustavo is a university professor, AI researcher, and international consultant on artificial intelligence. He brings the academic rigor and industry perspective that keep our wilder ideas grounded in reality. He's also the one who wrote our first tagline, during a late-night brainstorming session that produced more laughs than viable slogans.

Luiz has spent years building games across multiple startups and today works at one of the largest VR studios in the world — a Brazilian one, at that — shipping titles for major platforms. He splits his time between São Paulo and Portugal and brings a perspective the rest of us lack: one shaped by working at the frontier of how humans interact with technology. He's also the group's resident philosopher, the one who asks uncomfortable questions about where all of this is heading.

Bruno is an enterprise architect and a university professor — the guy who consistently scored top marks back in college and never really stopped teaching. He deals with the messy reality of deploying AI in large organizations. While
we debate theoretical architectures, Bruno is the one who has to make things work inside corporate firewalls, with real compliance requirements and developers who think "prompt engineering" means asking ChatGPT nicely.

Alex rounds out the group. He's a serial startup veteran and co-founder of an international backend-as-a-service platform. Less vocal in the chat but always lurking, running experiments on Linux boxes and occasionally surfacing with something that makes everyone stop and pay attention.

How "AIpster" Was Born

For the first two months, the group was called "MAKK HH 2023" — the remnants of its happy hour origins. As the conversations grew more serious, someone renamed it to "MAKK Empreendimentos 2024," which sounded appropriately ambitious and vaguely corporate.

Then, during a video call in February 2024, someone suggested "IApster" — a mashup of "IA" (the Portuguese abbreviation for artificial intelligence) and "hipster." Within minutes, Guilherme flipped it to "AIpster," because the English version sounded better and, let's be honest, hipsters don't acknowledge their own hipsterdom.

We asked one of our AI bots — yes, by that point we had already built bots inside the group — what it thought of the name. It said something diplomatic about brand identity and target audiences. We ignored the caveats and registered aipster.com the same week.

A tagline followed: "An AI-focused think tank applying cutting-edge technology to create ready-to-use AI PoCs and products." It was aspirational. It still is.

What We Actually Do

Mostly, we argue.

We argue about whether local models can compete with cloud APIs. We argue about whether AI agents should be given autonomy or kept on a short leash. We argue about token costs, model routing strategies, fine-tuning approaches, and whether the latest benchmark results mean anything at all.

But we also build. Over the past two years, we've collectively:

• Built and iterated on WhatsApp bots powered by local LLMs, complete with multi-agent architectures, translation layers, and more system prompt revisions than we care to admit.
• Experimented with AI-assisted coding workflows, testing everything from Claude Code to Cursor to BMAD, and developing opinions (strong ones) about when AI writes better code and when it confidently writes garbage.
• Trained custom ML models for financial forecasting, including ensemble methods, genetic algorithm hyperparameter tuning, and hierarchical fine-tuning pipelines.
• Explored model routing — the idea that different models excel at different tasks, and that a cheap flash model can handle pre-analysis while you reserve the expensive model for the hard questions.
• Debated the bigger picture: AI governance, the future of work, surveillance, the economics of GPU compute, and whether humanity is building something it can't control.

All of this happened in a WhatsApp group, scattered between memes, complaints about home renovation, and the occasional existential crisis.

Why a Blog

The conversations we have are messy, unfiltered, and often surprisingly deep. We disagree publicly and change our minds out loud. We share wins and failures with equal enthusiasm.

We realized that some of what we discuss might be useful to others — not because we have all the answers, but because we're asking the questions from a particular vantage point. We're not researchers at a lab. We're not VCs writing thought leadership. We're practitioners — people who pay their own cloud bills, manage their own servers, and deal with the gap between what AI promises and what it actually delivers on a Tuesday afternoon.

This blog will be a curated window into those conversations. We'll publish articles born from our discussions: practical insights, experiments, opinions, and the occasional rant. Some posts will be technical. Some will be philosophical. Most will be somewhere in between.

We won't pretend to be objective. We have biases — toward pragmatism, toward self-hosted solutions, toward keeping humans in the loop. We'll be upfront about them.

What to Expect

If you're interested in AI from the perspective of people who actually use it daily — not to pitch investors, but to solve their own problems — you might find something here worth reading.

Topics we'll cover include:

• AI coding tools in practice — what works, what doesn't, and what costs more than it should
• Local vs. cloud AI — the real tradeoffs, not the marketing materials
• Model routing and orchestration — how to use the right model for the right job
• ML for the real world — lessons from building models that have to make money, not just hit benchmarks
• The human side of AI — what changes when your tools start talking back

We're six friends who met in a computer lab in São Paulo a quarter century ago. The technology has changed beyond recognition since then, but the conversations haven't changed much at all: What can we build? How does it work? What could go wrong?

Welcome to AIpster. The think tank that started as a happy hour.

AIpster is an independent AI think tank based in São Paulo, Brazil. Follow us for practical insights on artificial intelligence from practitioners who build, break, and debate AI every day.