Truw

A piece of news lands in a chat. It reads true. Someone forwards it. By the time anyone checks it, the original message has already shaped what dozens of people believe.

We live in a moment where it's harder than ever to tell what's true. AI-generated videos, an election year and headlines engineered for engagement all pile on at once. Finding a clean read on what actually happened means hunting through ads, paywalls and outlets owned by groups with their own political interests.

The brief was simple. Make news verification fast, trustworthy and accessible to everyone, on every channel where the news already arrives. That principle shaped every decision, from the architecture down to the way a verdict is rendered on screen.

The project started as "Brasil Fatos", a web-first concept aimed at clean, ad-free news. The first board of flows mapped a reading experience around verified articles.

From there, the focus shifted to proactive verification: not just delivering verified news, but letting anyone verify what arrived in their inbox, group or feed. The audience model split into Reader, Specialist and Auditor / Journalist to anchor the freemium and pro tracks.

As Design Engineer I owned the interface end to end, prototyping and testing in short cycles with Claude Code inside Zed. Alongside developers, I worked on fine-tuning and RAG applied to fact-checking so the verdict was grounded in sources, not just confidence.

A dark, terminal-influenced surface, set entirely in JetBrains Mono for headlines, meta and verdicts; signal colors reserved for verdicts (green for verified, amber for doubtful, red for false).

Wurt, the brand character, was designed together with Hoff Research: a raven that watches before it judges. It's the voice of the verifier across the product.

The choice of JetBrains Mono wasn't only an aesthetic one. Reading a Truw verdict has more in common with a code review than with reading a magazine: the eye scans sources, hashes, dates and scores looking for the one character that's off.

The typeface was designed for that kind of attention, and it helps the reader trust what they're looking at.

The core loop is short: open the app, drop in what you want to check (text, voice, image or live camera), send. Within minutes Truw returns a verdict with credibility score (0 to 100), sources and the political spectrum of where that story is being told.

No paywall, no ads. The verified news arrives as a card to read, save or share, in a reading flow built for the story, not for the algorithm.

Most fact-checkers return a thumbs up or down. Truw returns context: who is amplifying the claim, where it sits on the political spectrum, and what the same story looks like through a different lens.

Four features carry that idea: the political spectrum bar on every card, Espectro (a quick read of where the reader sits politically, so the spectrum bar means something), Filtro de Bolha (read across ideologies on purpose), and multi-channel verification including a WhatsApp bot.

Truw ships on iOS, Android and a WhatsApp bot. The verification pipeline runs RAG on top of fine-tuned LLMs for fact-checking, returning the credibility score together with the sources that justify it.

The app isn't live yet. The landing page is up, and the iOS build was pulled from the App Store to make room for the next version with the updated verification model. Android and the WhatsApp bot follow the same cycle.

What this work proves: it's possible to design and build a multi-channel AI product end to end, in short cycles, without sacrificing visual quality or the transparency of the model behind it.