Why We Are Helping Build Guardião

Rita Barbosa’s grandmother got scammed. A phone call, a convincing story about undrinkable water, and a sale of four filters at a thousand euros each. This may seem expensive, but when the safety of drinking water is concerned, you spare no expense.

Except for the inconvenient fact that these products are worth about twenty euros, at best.

Losing three or four thousand euros was bad enough, but the consequences of breaching her trust were worse. This was a woman who had spent her life taking care of everyone around her, and now she was afraid of her own phone. She stopped answering calls and using the device altogether. She still needed her phone, but she didn’t trust it anymore. The loss of autonomy and confidence was infuriating to her entire family.

This happened right before Rita started her master’s in AI applied to medicine at the Hamburg University of Technology. She realized that recent advances in on-device language models had made something possible that genuinely wasn’t feasible six months earlier: she could, by herself, devise an app that could intercept scam calls and messages in real time, analyze them using AI, and block them before the person ever knows they happened. And this could be done without sending a single byte of data off the device.

She built a working prototype in 48 hours at the Cursor AI Hackathon in Hamburg — the largest AI hackathon in Europe, with over 400 participants from 48 countries — and won first place and a $70,000 prize. During her presentation, Rita noticed the audience wasn’t just evaluating her software architecture. They were recognizing the problem as their own. Almost everyone has a grandmother, an uncle, someone vulnerable.

Her brother Tiago saw the project and knew it had to come to Portugal. Tiago had already founded Three Sigma, a company in the fintech and Web3 space, and understood what it takes to build an actual product, robust enough to enter the market and become useful. Together, they created the startup behind Guardião.

Are you already sold on their vision by this point? We were.

Subvisual is joining Guardião as a development partner, investor, and venture partner.

As a development partner, we’re embedded with the team, accelerating the build across the full product surface. We’ve spent over fourteen years building products that do what Guardião needs to do. Reach enormous numbers of people (millions, if this is to work like we hope it will) — from Coverflex, the benefits platform that changed how hundreds of thousands of workers receive and manage their benefits, to StayAway Covid, the contact tracing app that reached millions of phones when the country needed reliable technology most. We know what it takes to go from prototype to product at scale, and we know the specific challenges of building consumer software in Portugal: accessibility, trust, and the gap between a working demo and something people depend on every day.

As an investor, we’re dedicating our team and resources to the mission. And as a venture partner, we’re bringing everything we’ve learned from over a decade of building and scaling companies to every strategic decision — product, go-to-market, hiring, the lot.

Most mobile security solutions work the same way: they detect something suspicious and then alert the user, leaving the actual decision-making up to them. That sounds reasonable until you think about who Guardião is actually for. As Rita puts it, the standard approach transfers responsibility to the person least equipped to handle it.

Guardião does the opposite. The AI runs entirely on-device, analyzes calls and messages before they reach the user, and blocks anything that matches scam patterns very quickly. It goes well beyond checking databases of known numbers. It detects manipulation tactics: artificial urgency, isolation pressure, emotional exploitation, and the specific conversational patterns that scammers use. If it’s a scam, the phone never rings. The message never arrives. It’s seamless.

We are building Guardião for the people who need it most, and that commitment is what shapes every technical decision we make. Therefore, Guardião will optimize for old phones since a lot of their intended audience uses cheap Android handsets with limited memory, slow processors, and even outdated software that someone’s grandchild set up for them at some point.

Privacy is equally non-negotiable. All processing stays on the device. Rita’s grandmother was a private person. She wouldn’t have accepted her conversations being sent to company servers, and neither would most of the people Guardião is designed to protect. There’s a family dashboard so the person who installed the app — typically a son or daughter — can see that threats were blocked, without ever hearing the conversations. Protection must never mean surveillance.

One more layer of complexity: scams in Portugal aren’t the same as scams in Germany or the United States. They use different tactics and appeal to different fears. The AI is being trained specifically on Portuguese fraud patterns, including cultural and linguistic nuances that make local scams effective.

We’re not the only ones who recognized what Guardião could become. The PSP (Portugal’s police force) signed a collaboration protocol with the project in March. Ricardo Toscano, who leads the cybercrime unit at PSP’s Department of Criminal Investigation, reached out to Rita’s team as soon as the hackathon win made the news. The alignment was immediate: the police deal with the consequences of these scams every day through programs like Apoio 65 and Escola Segura, and they saw a chance to help prevent them at the source.

For the product, the PSP’s contribution is criminal intelligence — known scam patterns, flagged numbers, operational insight into how fraud actually evolves in Portugal. They’re actively helping identify new methods so the AI can stay ahead of the curve. Longer term, there’s a vision for Guardião to serve as a bridge between citizens and the police: a way to report scams and get directed to the right support channels, directly from the app.

Nova FCT, where Rita completed her degree in Computer Engineering, has also joined as a partner. People involved with APAV — Portugal’s victim support association — will help with the testing phase, with people from their helplines contributing feedback based on the cases they handle daily. Google and OpenAI have provided technology credits to support the development.

Guardião is currently in internal testing. The next phase expands to roughly a hundred users with relevant expertise, including contacts from the PSP and APAV. From there, the plan is to scale through cohorts, going from a thousand users, then two thousand, then wider, before opening to everyone on the waiting list.

The initial release covers calls and SMS. Email and WhatsApp protection are on the roadmap. The PSP has also signaled interest in developing a broader digital safety toolkit in collaboration with Guardião, covering not just phone scams but email verification and fraudulent e-commerce sites.

Not every project we take on has this kind of direct, personal impact on people’s lives. When Rita tested the app against the exact scam that hit her grandmother, it caught it immediately.

This is why we build.

If you want to join the waiting list or learn more about Guardião, visit stopburlas.pt or give us a follow @guardiao.app.pt on Instagram and Facebook.

If you’re interested in how we build and invest in products like this, talk to us.