Inspiration
Every 30 seconds, a piece of misinformation is shared somewhere in the world. We have watched it cause real harm — vaccine hesitancy that killed people, election lies that toppled democracies, climate denial that delayed action on the planet's biggest crisis. In 2025 and 2026, AI-generated deepfakes and synthetic news articles made the problem exponentially worse. Billions of people scroll through content daily with no tool to help them tell real from fake. We are high schoolers who got angry about that. So we built something.
What it does
TruthLens is a free, open-source misinformation detector for anyone on the planet. Paste a headline, a WhatsApp forward, a tweet, or an article snippet in any language — and within seconds you get a credibility score from 0 to 100, a verdict (Likely False, Unverified, or Likely Credible), plain-English red flag explanations like "no source cited", "emotional manipulation", or "contradicts established science", and a breakdown of exactly why the content scored the way it did. It does not just label misinformation. It teaches critical thinking — so next time, you spot it yourself.
How we built it
We built TruthLens full-stack in under 48 hours. The frontend is React with Tailwind CSS, the backend is Node.js and Express, and the intelligence comes from Anthropic's Claude API acting as a computational linguistics expert. We engineered a structured NLP prompt that instructs Claude to detect specific linguistic red flags — vague attribution, sensationalist word choice, absent citations, emotional framing, logical fallacies, and contradiction with peer-reviewed sources. Claude returns structured JSON which we parse and render into a clean, color-coded results card. Every design decision was made with one user in mind: someone with no media literacy training, reading news on a cheap smartphone.
Challenges we ran into
Getting consistent structured output from a large language model was harder than we expected. Early versions of our Claude prompt returned markdown-wrapped JSON, inconsistent field names, and occasionally broke our parser entirely. We iterated through a dozen prompt versions before landing on one that returned clean, reliable JSON every single time. We also struggled with nuance — satire, opinion pieces, and heavily biased-but-not-false content kept confusing our scoring. We solved this by adding an "Unverified" middle tier and tuning the prompt to distinguish between factual falsehood and editorial bias. CORS configuration between our React frontend and Express backend also cost us more time than we'd like to admit.
Accomplishments that we're proud of
We are proud that TruthLens works on content in multiple languages — Hindi, Spanish, French, and more — because misinformation is not an English-only problem. We are proud of the explanation layer: instead of just giving a score, we surface the exact linguistic signals that triggered suspicion, which has real educational value. Most of all, we are proud that we built something genuinely useful in 48 hours that addresses one of the most important problems in the world right now.
What we learned
We learned that prompt engineering is itself a form of computational linguistics. Choosing the right words to instruct a language model — the precision, the constraints, the output format — is exactly the same discipline as designing a grammar or annotation scheme in NLP research. We also learned that misinformation detection is not a binary problem. Real-world content lives on a spectrum, and any tool that pretends otherwise will mislead people just as badly as the fake news it is trying to fight.
What's next for TruthLens
We want to build a browser extension so TruthLens works inline as you browse — no copy-pasting needed. We want to add source credibility checking by cross-referencing claims against Google's Fact Check Tools API and live news databases. We want to support image and video analysis to tackle deepfakes. And long-term, we want to partner with schools across India, Africa, and Southeast Asia — the regions most vulnerable to WhatsApp misinformation — to integrate TruthLens as a free media literacy tool in classrooms. The fight against misinformation is generational. We are just getting started.
Log in or sign up for Devpost to join the conversation.