Developed under the PURE program at Sabancı University, this VR SDK aims to enhance accessibility for low vision individuals. Featured a comprehensive two-phase user study to ensure effectiveness and ease of use while using.
Developed under the PURE program at SUNUM (Sabancı University Nanotechnology Research and Application Center ). This project developed a highly customizable electrospinning device that uses Arduino and stepper motors to precisely control several key parameters, including drum-pump distance and pump rotation speed. The device is designed to advance research in nanofiber production technology by facilitating detailed studies on fiber diameter, strength, and interconnectedness. Features include dual adjustable nozzles, variable collector drum configurations, and a robust mechanical framework that ensures precise alignment and handling of electrostatically spun fibers. The integration of software and mechanical components allows for dynamic adjustments to the spinning process, optimizing the production of nanofibers for various applications.
This project uses deep reinforcement learning with curriculum learning to guide AI agents through increasingly challenging scenarios, helping them master game tasks in stages. It also includes imitation learning, where agents learn strategies by watching human gameplay. These player-like tactics are compared to fully self-taught strategies, highlighting how agents perform with human guidance versus independent learning. Built in Unity, the game environment provides a flexible setup for testing AI behaviors. The game's foundation was created by my sister under my guidance, with a tutorial, and as a surprise for her, I trained AI to play it ;)
This project explores the potential of chaining GAN models, focusing on using BlendGAN and InterFaceGAN, to enhance the flexibility and precision of stylized semantic face editing. In this approach, a SVM trained on latent vectors and corresponding attribute scores, along with a trained latent encoder, establishes boundaries for semantic attributes, such as age, gender, and expressions. These SVM-trained boundaries and directions enable fine-grained image editing by performing vector operations within StyleGAN’s latent space, enhancing control over specific facial features. Following this, BlendGAN transforms the semantically edited image by blending it into various artistic styles, using an encoder-decoder setup for style transfer and the AAHQ dataset for diverse stylistic references. This chain of GANs highlights the capability to preserve the semantic integrity of edits introduced in the input's latent space, with attention to maintaining high fidelity in attribute transfer through the entire stylization process.
Can Wait is a 2D roguelike indie game developed in Unity, during an internship at Vourlabs, available on itch.io. Players navigate through procedurally generated maps crafted with the wave function collapse, engaging in battles against monsters and collecting byte coins. These coins are used to purchase upgrades and enhance abilities. The game supports both local and networked co-op play, allowing friends to play together. Featuring different coloured robot characters and a variety of combat scenarios. A demo is currently available, on itch.io and the code is available on GitHub.
The SU_Govern project presents a recursive hierarchical DAO (decentralized autonomous organization) designed for transferring structured organizations to blockchain based active governance. Deployed on the Polygon blockchain, it enables transparent and dynamic governance with proposal management, token lending, organizational and individual voting, leveraging ERC-20 tokens for membership and incentives. This recursive system allows interconnected DAO governance, providing administrative privileges that enhance depth and flexibility in decision-making. For more information, you can visit the project on GitHub.
We have developed a biometric multifactor authentication security system that integrates IoT and edge server technologies. This system enhances facility security by utilizing facial recognition and voice verification methods. It starts by detecting an individual's presence with an ultrasonic sensor. When someone approaches, the system takes their photo, compares it against doors local database using a VGG face encoder model, and identifies if they are recognized. If the person is known, the system then prompts them to verbally provide a password. This speech is recorded and processed by OpenAI's Whisper and a voice encoder model, ensuring accurate recognition across multiple languages. Instead of physically unlocking a door, we use LED lights to indicate access status. This project prioritizes security and data integrity by storing sensitive information locally on the edge devices and performing the computation on cloud.
The Corona Hunter is an open-world FPS game using C++ and OpenGL. This game has a custom-built engine and four diverse biomes. Players navigate a procedurally generated world, targeting coronavirus shaped enemies with vaccine-bullets. Technical highlights include rigid body dynamics for natural movement, AABB and custom collider collision detection, a chunk system for optimized scene graph performance, LOD and culling systems for efficient rendering, and comprehensive lighting, shading and texture mapping techniques.
The Olympus Bookstore is an e-commerce platform for selling books, with both a website and mobile app that support user account creation, item listing, and purchases. It features advanced search and filtering, product pages with detailed information, and management interfaces for product and sales managers to handle categories, reviews, and promotions. Developed by a team of six using Scrum methodologies, where i was the scrum master , the backend is built on PostgreSQL and Python Flask, while the frontend utilizes React for the web and Flutter for mobile. The project is fully documented on GitHub, showcasing the application's capabilities and underlying technologies.
This CPU ray tracer combines techniques from "Ray Tracing in One Weekend," "Ray Tracing: The Next Week," and "Ray Tracing: The Rest of Your Life." It supports complex lighting and realistic materials, including diffuse, specular, and refractive properties, to create photorealistic images. This project also includes a mesh parser to import .obj files.
Currently pursuing a Master's degree in Artificial Intelligence at Utrecht University.
GPA: 3.89/4.00, (Summa Cum Laude).
Ranked Top 10 of the graduating class.
Recipient of Honors and listed on the Dean's List for 6 contiguous terms.
Ranked second in graduating class.