Projects

End-to-end full-stack Generative AI system, built as a university course term project, running entirely on local GPU hardware to transform raw product photos into professional studio visuals with zero cloud dependency.

• •Integrated four coordinated model components — a quantized FP8 UNET diffusion core, a 7B-parameter vision-language CLIP text encoder, a VAE, and a Lightning LoRA adapter — applying Prompt Engineering to preserve product identity across multiple camera angles.
• •Designed three distinct camera-angle prompt pipelines (hero shot, three-quarter profile, isometric view) with tailored positive/negative prompting to eliminate studio equipment, human hands, and artifacts from generated output.
• •Architected a React and Vite frontend utilizing WebSockets for bidirectional, real-time progress logging.

Computer Engineering graduation project: an AI analysis system that detects game tiles on a physical rack, classifies them, and mathematically models game rules to recommend optimal moves.

• •Engineered a custom dataset from scratch — 4,432 tiles labeled across 219 rack photographs via Roboflow for localization, plus a custom keyboard-driven rapid annotation tool to classify tiles into 55 distinct classes for the classification stage.
• •Implemented a Two-Stage Hybrid Pipeline utilizing YOLOv8-Nano (99.4% mAP50, 41ms inference) for localization and ResNet-18 (97.26% accuracy, 2.88ms inference) for classification, selected after benchmarking against EfficientNet-B0, ViT-B/16, and LightGBM.
• •Modeled complex combinatorial problems utilizing Depth-First Search (DFS) and Memoization to solve game strategies, executing in under 10ms in-browser.
• •Architected a Client-Direct structure using an isolated Flask backend, eliminating serverless cold-start latency and achieving a 150-250ms end-to-end real-time response via a Next.js interface.
• •Built an interactive correction UI allowing users to relabel misclassified tiles in real time, feeding corrected samples back into the dataset as a continuous-improvement loop.

Fully containerized distributed backend system for managing customers, orders, and products built with Java, Spring Boot, and Spring Cloud.

• •Developed a Spring Boot backend for managing customers, orders, and products in a structured web application.
• •Architected a fully containerized distributed system utilizing Java, Spring Boot, and Spring Cloud.
• •Designed the system using both monolithic and microservices architectures, leveraging OpenFeign for inter-service communication.
• •Implemented dynamic service discovery via Eureka Server and centralized routing utilizing an API Gateway.
• •Built the project as a RESTful API, implementing CRUD operations with proper request validation, error handling, and standardized response structures.
• •Implemented MVC and DTO patterns for clean architecture and efficient data handling.
• •Integrated MySQL as the relational database for data storage and management.
• •Used Docker and Docker Compose for containerization and deployment.

Co-authored research paper (Marmara University, Dept. of Computer Engineering) on detecting suicide-related content in social media posts using deep learning and NLP, with comparative results across two transformer architectures.

• •Authored a research paper on detecting suicidal intent in social media posts using deep learning models.
• •Implemented and compared two unused models in literature, achieving 97% accuracy in both models.
• •Conducted data preprocessing, feature extraction, and model evaluation to improve classification performance.

A fully isolated cybersecurity lab simulating a WannaCry attack to analyze ransomware behavior and defense strategies.

• •Simulated a WannaCry ransomware attack in a fully isolated virtual environment to study infection patterns and defense strategies.
• •Configured Windows 10 and REMnux virtual machines on a host-only network to ensure total isolation from production systems.
• •Deployed INetSim to safely emulate internet services, enabling observation of malware's outbound communication attempts.
• •Monitored system and network activity using Wireshark, tcpdump, Procmon, Process Explorer, and Regshot to capture behavioral indicators.
• •Performed static analysis with PeStudio to examine malware binaries, structure, and imported libraries.
• •Analyzed file encryption processes, registry modifications, and lateral movement techniques commonly used by WannaCry.

Optimization modeling project simulating constraint-based decision support for real-world logistics bottlenecks and dynamic vehicle capacity constraints.

• •Modeled real-world logistics bottlenecks and dynamic vehicle capacity constraints utilizing Mixed Integer Linear Programming (MILP).
• •Processed large-scale operational datasets and executed algorithmic optimization leveraging Python, Pyomo, and Gurobi solvers to simulate constraint-based decision support.

A Pacman themed multiplayer game made with unity.

• •Developed a multiplayer version of the classic Pac-Man game using C# and Unity.
• •Implemented real-time player-controlled ghosts with Photon for networking and multiplayer synchronization.
• •Designed game mechanics to maintain the original Pac-Man experience while enhancing interactivity.
• •Focused on smooth network communication, responsive controls, and an engaging multiplayer experience.

Simple Car Game Coded With Assembly

• •Programmed a hardware-near logic control simulation utilizing 8086 Assembly and Emu8086, demonstrating computational fundamentals and register-level programming.

An application that helps you manage your weekly courses.

• •Developed a GUI-based scheduling system utilizing Java and JavaFX featuring multi-user configuration logic.
• •Easy to use, user friendly interface.
• •8 Course time slots for 7 days of the week.
• •Choose your course type.
• •Different syllabi for different accounts.

A website for observing a scale model of the solar system.

• •Modeled a 3D simulation utilizing JavaScript and Three.js, implementing 3D rendering, interactive camera controls, and animation loop modeling.
• •Realistic 3D models of all planets in our solar system.
• •Interactive camera controls for exploring the solar system.
• •Planet trails showing orbital paths.