What it does Tinderbox is a full desktop Tinder interface built on the Play! 2.2 framework. It runs a local server at localhost:9000, authenticates via Facebook access tokens, and recreates most core Tinder functionality in a browser. The twist: background jobs using Akka and Spark handle facial detection, while a bot manages initial conversations.

The interesting bit The project uses Eigenfaces—grayscale, normalized facial pixel models—to learn your “type” from past likes and dislikes. It’s deliberately elementary: eigenvector values from yes/no face models determine new swipes. For messaging, a decision tree routes replies based on positive or negative sentiment detected via Stanford NLP, then hands off to you when the conversation escapes the script.

Key highlights

• Eigenfaces-based auto-swiping with a “Clear Models” reset button when its taste goes awry
• Sentiment-driven chatbot with hardcoded message trees in FunMessages.scala
• Packaged as an OSX .app launcher plus shell scripts for OSX/Linux (Windows untested)
• Requires manual Facebook token extraction with a one-second copy-paste window before FB obscures it
• Non-commercial CC license; author explicitly warns against spamming and disclaims any affiliation with Tinder

Caveats

• Sunsetted and unsupported: the original author abandoned the project and moved to “Bernie AI”
• API fragility: any Tinder API change will break the app
• Immediate messaging risk: the bot starts messaging all contacts as soon as launched, before you can customize messages
• Elementary ML: the author admits the facial analysis “could use some work” and suggests combining Eigenfaces with other methods

Verdict Worth a look for ML historians or anyone studying early-2010s bot behavior and crude facial recognition pipelines. Not for actual dating, production use, or anyone hoping for maintained code. The Eigenfaces implementation is more instructive than effective.