Aesthify - AI-Powered Design Intelligence

The Problem

Designers talk about balance, symmetry, and contrast, but there's no structured way to quantify these principles or connect them to how users actually perceive aesthetics. Everyone relies on intuition without understanding what users actually prefer.

What I Built

The technical framework

I used YOLOv8 for object detection in interior design layouts, then built rule-based scoring for 7 design principles: balance, proportion, symmetry, simplicity, harmony, contrast, and unity. I chose rule-based over machine learning because I wanted the system to be explainable and grounded in actual design theory, not a black box.

Built a Flask interface that shows real-time scoring with bounding box visualizations. The architecture supports both local (YOLOv8) and cloud (Roboflow) detection.

The user study

I recruited 101 participants to rate 8 interior design images on a 1-9 scale. Collected ratings, emotion tags, style preferences, and demographics. Then analyzed patterns across age groups, professions, countries, and design styles.

Key Findings

What actually predicts user preference:

Simplicity: r = 0.68 (strongest predictor)
Contrast: r = 0.56
Unity: r = 0.40
Symmetry: r = -0.60 (negative correlation, users rated symmetric layouts lower)

Demographic patterns:

18-25 year-olds consistently preferred minimalist, modern designs
Design and psychology backgrounds gave higher ratings than engineering or medical
Cultural variation was significant: Netherlands/UAE preferred traditional, India/Canada leaned modern

Three distinct user groups emerged:

44%: Calm, Minimalist, Elegant
31%: Energetic, Elegant, Cozy
26%: Calm, Elegant, Dull

Impact

Open-sourced on GitHub and currently used by design research scholars for data-driven decisions. The study proved empirically that aesthetic perception varies significantly by demographics and culture. The negative correlation with symmetry challenged traditional design assumptions.

What I Learned

The tool is only valuable because of the user study. Without connecting computation to actual perception, it's just another aesthetic scorer. The wide variance in ratings for identical layouts proves that aesthetics is inherently subjective and culturally shaped. You can't design for "everyone" because everyone sees things differently.