An AI-powered web app that uses Retrieval-Augmented Generation (RAG) to verify news articles and headlines.
Verify News
AI Experiment
Challenge
I wanted to move beyond a static Figma prototype and build a fully functional tool that doesn’t just label a headline as fake, but explains why.
Because live AI models can still occasionally hallucinate, this project serves as an active experiment testing the limits of design and prompt engineering.
Building with Cursor & AI
Using Cursor and Gemini Pro as co-pilots, I built a Next.js backend that scrapes web articles and verifies claims. The architecture has been built with Llama 3 model and Brave Search API to feed the app real-time web data.
To reduce hallucinations, I added strict prompting rules: the system automatically extracts publication dates for accurate temporal context, and restricts AI outputs to 15-word bullets with source citations.
Designing a Frictionless UX
To keep the complex data digestible, I designed a clean React/Tailwind UI that dynamically shifts into a masonry grid based on the volume of evidence.
I also turned technical constraints into UX features. I designed a custom error state for API limits, and a zero-click reset that instantly clears the screen when a new link is pasted.
