SCREENBot APP

Shazam for Screenshots. Clean. Glean. Extract.

Comment

=============================== supercreativepeople.com/screenbot

screenbotapp@gmail.com

INSPIRATION

I'm an Emmy Award-winning creative technologist with 30 years in media - TMZ, MTV, CBS, Google/Waymo - and like everyone else, my iPhone camera roll is a disaster. Thousands of screenshots: songs I heard on the radio and couldn't Shazam in the car, restaurants I saved from Instagram, movies that looked good in someone's Story, products that caught my eye. All of it piled up, unsearchable, forgotten.

Shazam solved "I can't name this song." I wanted to solve "I can't find anything in my screenshots." That became SCREENBot.

WHAT IT DOES

SCREENBot scans your iPhone Screenshots album, uses AI vision to classify and extract metadata from each image, and routes you directly to action - one tap to Spotify, Yelp, TMDB, Ticketmaster, Amazon.

Three stages. One tap: CLEAN - Screenshots offloaded, device storage reclaimed. GLEAN - OpenAI Vision reads each image and extracts structured data: song titles, restaurant names, movie info, product details, event info. EXTRACT - Results arrive as action chips. Tap once to open Spotify, look up a restaurant on Yelp, or buy tickets on Ticketmaster.

90%+ classification accuracy across 8 smart categories: Music, Movies & TV, Dining, Bars, Events, Jobs, Shopping, and Other.

The UX is gamified by design. While the AI works, the SCREENBot mascot animates through a "wipe" loop - physically cleaning the screen while the status cycles through Cleaning... Gleaning... Extracting... The wait becomes the experience.

HOW I BUILT IT

Built using an AI-augmented development stack from concept to App Store submission in a single sprint -zero traditional engineering headcount.

Frontend: React Native (Expo) - universal iPhone + iPad Backend: Python / Flask on Google Cloud Run AI: OpenAI GPT-4o-mini with vision - custom system prompt tuned for real-world screenshot noise Enrichment: iTunes Search API + TMDB for metadata Payments: RevenueCat + Apple StoreKit 2 Analytics: Mixpanel (App_Opened, Screenshot_Uploaded, Analysis_Complete)

CHALLENGES I RAN INTO

Classification accuracy was the core technical challenge. Real-world screenshots are noisy - TikTok comment sections look like movie listings, car stereo Bluetooth lock screens get misclassified, social layout variation is huge. Getting to 90%+ required iterating the GPT-4o-mini system prompt through multiple classification cycles across a 118-image real-world test set.

Cloud Run introduced an early infrastructure problem: Flask tried to write to ./uploads/, which doesn't exist in a stateless container. Switching to /tmp/uploads resolved it and forced cleaner serverless architecture.

The App Store gauntlet is its own challenge. RevenueCat cannot serve offerings until ASC products pass their first review - meaning you ship before you can fully test payments in production. First submission rejected on two guidelines (photo library purpose string, IAP access path). Both resolved and resubmitted.

ACCOMPLISHMENTS I'M PROUD OF

  • 90%+ classification accuracy on 118-image real-world test set
  • Full Clean-Glean-Extract loop operational end-to-end
  • Universal iPhone + iPad app with gamified UX (the Wipe Routine)
  • In-app purchases live: Free (3 scans/month), Pro Monthly ($4.99), Pro Annual ($39.99)
  • Mixpanel analytics wired end-to-end
  • Legal infrastructure live at supercreativepeople.com/screenbot
  • Concept to App Store in a single sprint with zero engineering headcount

WHAT I LEARNED

Building with AI APIs in production is fundamentally a prompt engineering and data validation challenge, not just a coding problem. The gap between 70% and 90% accuracy came entirely from how the task was framed and what fallback logic was built around edge cases.

The "solo founder can't ship" myth is dead. With the right AI-augmented stack, a single operator can take a product from concept to a production-quality App Store submission on a sprint timeline. SCREENBot is the proof of concept.

WHAT'S NEXT FOR SCREENBOTAPP

  • Remove batch limits - full library processing, thousands of screenshots at once
  • Direct platform integrations that act, not just surface: build the Spotify playlist, make the OpenTable reservation
  • Android expansion
  • Desktop power mode

The bigger play: SCREENBot is a referral engine. No inventory. No allegiance. We send platforms warm, high-intent users. Every major platform respects the bot because it feeds them all.

Pre-seed: $150K-$250K. Investor access: supercreativepeople.com

Built With

Share this project:

Updates