Hello, I'm Emre

A full-stack multi-agent AI research platform that transforms academic papers into an interactive knowledge workspace. It coordinates six specialized AI agents to ingest, structure, map, critique, hypothesize, and write - generating concept graphs, structured wikis, gap analyses, scientific hypotheses, and research briefs from raw PDFs. Built to help researchers focus on strategic decisions rather than manual literature processing. Won 3rd Place + Google DeepMind Special Recognition at AI Agent Olympics, Milan AI Week 2026.

Developing a cutting-edge website powered by LLMs that provides a unique service for building custom microcourses. This platform will allow users to effortlessly design their own microcourses, tailored to their specific needs, through the advanced capabilities of LLM technology. Whether users wish to quickly refresh a topic they have previously learned or gain concise insights into a new subject, our AI will facilitate the process seamlessly. The AI will not only set up the course content according to user requirements but will also automatically generate essential learning tools such as Recall-Cards, Quizzes, and a Glossary. This comprehensive approach ensures a holistic learning experience, enabling users to maximize their understanding and retention of the subject matter efficiently. This project also embraces state-of-the-art, scientifically proven learning methods to maximize the learning experience. In summary, this LLM-powered website aims to revolutionize the way individuals generate and engage with educational content, offering a personalized, efficient, and highly effective learning solution.

Developed a Turkish-speaking, AI-powered call center assistant leveraging advanced Natural Language Processing (NLP) techniques, Large Language Models (LLMs), and real-time intent and sentiment analysis. Integrated context-aware dialogue management and agentic AI design to deliver user-focused, responsive customer support simulations. Optimized to reduce wait times and enhance satisfaction, the system was built using Python, machine learning frameworks, and cloud-based deployment tools, and achieved a Top 20 ranking in TEKNOFEST's Turkish NLP Scenario Category.

Corners are points where gray value intensity changes sharply in two or more directions, and they are essential for tasks like character recognition, graphic matching, motion detection, 3D modeling, and object tracking. Classical corner detection algorithms such as FAST and Harris are computationally efficient, but they often lack sufficient accuracy and repeatability. This study implements a Quantum Harris Corner Detection algorithm. To evaluate its performance, a dataset containing images of circles, rectangles, triangles, and squares is encoded using two quantum image models—FRQI and QPIE—for comparison. To further enhance the algorithm's output, classical post-processing is applied to each detected corner point.

Our aim was to explore how agentic AI systems could be applied to public feedback and sentiment analysis in Turkish. We built a prototype where a fine-tuned Mistral-7B (mistralai/Mistral-7B-v0.1) handled sentiment classification (positive, negative, neutral), while a LLaMA 3.1 8B model acted as an orchestrator with reasoning and tool-calling abilities. The Agno framework tied everything together in an agentic setup. Training and experimentation were carried out on High-Performance Computing (HPC) resources provided by TÜBİTAK, which enabled us to fine-tune and evaluate large models efficiently.

Developing a time management web application for Twinformatics aimed at enhancing user efficiency and productivity. The project involves designing and building a sophisticated platform to help users optimize their work schedules and boost individual productivity. Key features include streamlined task organization, efficient time allocation tools, and overall improvements in work-time management.

Developed a simple quantum game within the QJam 2023/24 framework, where the user controls a Mario character in a quantum superposition state. The character is quantum simulated, and the gameplay involves navigating obstacles—if the character hits one (i.e., gets observed), its superposition collapses. The project utilized technologies such as Python, Construct 2, and Qiskit.

Developed a hand gesture recognition application using Python, OpenCV, and a hand tracking module, based on a video tutorial and further enhanced independently. The project focused on gaining hands-on experience in real-time image processing and gesture recognition—key areas in computer vision. The application captures video through a webcam and uses the HandDetector module to detect and track hand movements. Core components included handling video input, implementing hand detection algorithms, and interpreting hand gestures.