Shipped a live, full-stack RAG tool on Vercel
Turned technical limits into UX: custom API-limit state
Reduced hallucinations with strict prompting and temporal context
AI Experiment
Verify News
An AI-powered web app that uses Retrieval-Augmented Generation (RAG) to verify news articles and headlines.
Platform
Web
Focus
Product design + build, prompt engineering
Tools
Cursor, Gemini, Llama 3, Brave API, Tailwind, Next.js, GitHub
Impact
The Challenge ⭐️
I wanted to move past a static Figma prototype.
So I built a working tool that explains its verdict instead of just labeling a headline fake.
Building with Cursor & AI
I used Cursor and Gemini Pro as co-pilots to build a Next.js backend that scrapes articles and verifies claims. Llama 3 and Brave Search API feed it real-time data.
To curb hallucinations, I added strict rules. First, the system extracts publication dates automatically for temporal context. Then it caps every AI output at 15-word bullets with source citations.
Designing a Frictionless UX
Dense data needs to stay digestible, so I designed a clean React/Tailwind UI.
It shifts into a masonry grid based on how much evidence comes back.
I also turned constraints into features: a custom error state handles API limits, and a zero-click reset clears the screen the moment you paste a new link.
