Hi, I'm Alper

Built AI-powered personal styling app solving daily outfit decisions for users. Developed complete fashion-tech ecosystem including "Gimme Looks" (AI outfit recommendations based on wardrobe, mood, and occasion), "Shopping Advisor" (real-time garment analysis with wardrobe compatibility matching), and "Planner" (trip and weekly outfit planning). Engineered computer vision pipeline with YOLO11 for clothing detection, Segment Anything Model for precise segmentation, and FashionCLIP-powered visual search across 100K+ products from partner APIs covering 6 European countries. Deployed scalable backend with FastAPI, Azure, Firebase and Open AI/Anthropic LLM APIs for personalized insights.

Pioneered hybrid Life Cycle Assessment-Agent Based Model integrating nonlinear multi-objective optimization to balance environmental sustainability with economic viability in dairy farming. Simulated 1,800+ Luxembourg farms using Java and Python, optimizing operations across conflicting constraints (carbon footprint reduction, profitability, regulatory compliance). Developed Django web platform serving 10 regional stakeholders with real-time farm performance analytics. Research contributions include novel approaches to soybean reduction in dairy cattle diet reducing CH4 emissions, biogas feedstock optimization, and farmer decision-making simulation. Published several papers, demonstrating quantifiable environmental impact. Collaborated with agriculture consultants, engineers, and farmer cooperatives translating complex models into actionable policy recommendations.

Developed unsupervised machine learning framework for real-time bearing fault prediction in streaming sensor data, addressing critical Industry 4.0 predictive maintenance challenges. Implemented and benchmarked four algorithms (Mahalanobis distance, Bayesian changepoint detection, SPLL, LSTM-Autoencoder) on IMS and XJTU-SY vibration datasets, achieving earlier fault detection with linear time complexity. Applied dimensionality reduction (PCA, t-SNE) and advanced signal processing (wavelet transforms) to extract features from streaming vibration data. Demonstrated Bayesian and Mahalanobis methods as optimal choices for IIoT deployment, enabling cloud-based monitoring frameworks that reduce continuous human supervision while maintaining high accuracy in detecting bearing degradation severity levels.

Interactive web application visualizing Luxembourg housing market data across communes from 2010-2024. Built an intuitive map-based interface allowing users to explore house and apartment prices by region, property type, and price percentiles. Implemented dynamic filtering controls and geographic visualization to help users understand real estate pricing trends across Luxembourg. The application provides a free, accessible tool for real estate research and market analysis with responsive design and optimized performance.