AI Taste App

A private dating experiment that learns your visual preferences from yes/no photo ratings and scores new portraits against that feedback.

Teach: Select a portrait and record a personal yes/no reaction.
Teach: Select a portrait and record a personal yes/no reaction.

About AI Taste App

AI Taste App started with a personal dating question: could an application learn which faces I find attractive from a series of simple yes/no reactions?

I built a browser application around that idea, with two modes: Teach and Predict. In Teach mode, you upload a portrait and respond with “Yay” or “Nay.” The application saves that feedback as a training example. In Predict mode, you submit another portrait and receive a score based on the patterns learned from your previous choices. Prediction keeps the existing training examples unchanged.

Behind the interface, the application detects and crops the face, then uses a pretrained SigLIP model to convert the image into a numerical representation called an embedding. A small personal classifier learns from those embeddings and your ratings. The pretrained model handles visual features, while the classifier adapts to your feedback.

The web application uses Laravel, Blade, JavaScript, and Tailwind CSS. A separate Python service built with FastAPI handles image processing and machine learning through MediaPipe and PyTorch. Saved feedback, image embeddings, and classifier weights connect the teaching and prediction stages across requests.

The engineering work involved bringing photo uploads, face detection, feedback collection, model training, and prediction together in a straightforward mobile interface. That makes the project a practical experiment in connecting a conventional web application to a personalized machine-learning service.

AI Taste App remains a private personal prototype. Its scores represent the model’s response to an individual’s visual preferences. The experiment explores attraction from photos; it does not establish relationship compatibility or a universal attractiveness rating.

Screenshots note: Illustrative interface screenshots using AI-generated portraits of fictional adults. Displayed scores are model-output examples, not accuracy measurements.

Release history

Version 1.0.0