Shipped a live AI product with a signed-off fictional persona
Designed a post-verdict engagement loop: archetype, score, recommendations.
AI Experiment
The Critic’s Darkroom
An AI-powered roast engine that judges your movie taste, scores you, assigns you a critic archetype, and reluctantly recommends what to watch next.
Platform
Web
Focus
Product Design + Build, Prompt Engineering
Tools
Claude Code, Llama 3, Next.js, Vercel, GitHub
Impact
The Challenge
Most AI demos are polite and forgettable.
I wanted to build one with enough personality that people screenshot the result and send it to a friend.
Building the Critic's Voice
I built the app in Next.js with Llama 3 powering the roasts.
The prompt architecture gives the critic a consistent persona: sharp, literate, and personally offended by your choices.
The Visual Rework
The UI was rebuilt directly in Cursor with Claude Code, pushing toward bold editorial typography and a darkroom-inspired palette anchored by a deep gold accent on black.
Poster art populates each card automatically when you type a title, so the input screen feels like a shelf, not a form.
After the Roast
After the roast, a Critical Archetype card assigns you a personality type with a score out of 100 and tips to raise it.
Then a “Reluctantly Recommends” section suggests films to redeem your taste.
Share and Save sit below the verdict and suggestions in case a user wants to keep them. 😀