/// AI App Case Study
Screen Time Roast
AI that roasts your digital habits
Category
AI App
Stack
Next.js, TypeScript, Tailwind CSS, OpenAI API
Year
2024
01 / The Context
Built With
Live At
https://screentimeroast.web.idAI that roasts your digital habits
The Screen Time report on my phone told me I'd spent 4 hours on social media in a single day. The app just... showed me the number. No reaction. No context. No judgment. That lack of judgment felt like a missed opportunity.
What if the app roasted you? Not meanly — but in the way a good friend would. Honest, a little brutal, and ultimately pushing you toward better habits while making you laugh about the current ones.
Screen Time Roast was built in 48 hours for a personal challenge. Upload your screen time, get AI-generated commentary that is equal parts savage and weirdly motivating.
02 / The Problem Space
Defining the core
user pain points.
01
Prompt Engineering for Humor
Getting an LLM to be consistently funny — not cringe, not offensive, not repetitive — is genuinely hard. Dozens of iterations of the system prompt before it hit the right tone.
02
Parsing Unstructured Screen Time Data
Screen Time exports aren't standardized. iOS formats differ from Android, and manual entry had to be an option. Building a flexible parser that handled edge cases took more time than the AI integration.
03
Going Viral Without a Budget
This was a "build it and share it" project. The product had to be immediately shareable — results needed to be copy-pasteable, meme-able, and funny enough to screenshot.
03 / Strategy & Execution
From hypothesis
to prototype.
2 Hours
One voice memo. One Notion doc. A tweet draft that never got sent. The concept was clear enough to start building immediately — a rare feeling that usually means you're onto something.
6 Hours
Iterated on the OpenAI system prompt ~40 times. Each iteration tested against 5 sample datasets. The final prompt is weirdly specific about tone, length, and roast intensity levels.
1.5 Days
Next.js + OpenAI API. File upload handler, screen time parser, streaming API response for the roast reveal effect. TypeScript for sanity.
4 Hours
Deployed to Vercel. Shared in 3 group chats. Had 200 uses by end of day. Someone posted it to Twitter. It had a moment.
04 / Impact & Learnings
Measuring
success.
800+
Roasts generated
In the first week.
40+
Prompt iterations
To get the tone right.
48h
Concept to launch
Including sleep.