SpotHopper is an intelligent iOS application for personalized venue recommendations, developed as part of my Bachelor's thesis on 'Data-Driven Personalization'. To effectively resolve the cold-start problem and maximize suggestion relevance, I engineered a hybrid recommender system that fuses content-based filtering with explicit user feedback.

The project represents a comprehensive full-stack implementation: the backend relies on Firebase for data management, while the frontend is built natively with SwiftUI. The core intelligence is a machine learning pipeline trained in Python (TensorFlow/Keras) on OpenStreetMap data. I optimized the model using TensorFlow Lite, enabling latency-free on-device inference without permanent server dependency.

A central design feature is the focus on conscious exploration: to break habitual patterns, users receive only a single, highly tailored suggestion every 24 hours. In evaluation, this hybrid architecture achieved a 175% increase in prediction accuracy (Precision@1) compared to purely preference-based baselines. This project demonstrates the seamless integration of rigorous academic research into modern, user-centric mobile development.