I'm a Computer Engineering student at İzmir Institute of Technology (IZTECH) interested in software engineering, machine learning, natural language processing, and intelligent software systems.
I enjoy building practical software applications as well as experimenting with machine learning models on real-world datasets.
My current work spans areas such as time-series modeling, large language models, healthcare AI, data mining, and cross-platform application development.
- Deep learning with PyTorch
- Machine learning with Scikit-learn and XGBoost
- Hyperparameter optimization with Optuna
- Transformer-based architectures
- Time-series modeling
- Natural Language Processing
- Large Language Models
- Prompt engineering
- Ensemble learning
- Model evaluation and experimental analysis
- Cross-platform application development with Flutter & Dart
- Responsive web development with Next.js
- Real-time applications with Firebase & Firestore
- Object-oriented software development with Java
- Connected-device development with Bluetooth Low Energy (BLE)
A machine learning project for predicting ICU length of stay from longitudinal clinical time-series data derived from MIMIC-IV.
The project compares a Temporal Transformer with several machine learning and neural-network baselines.
- Temporal Transformer architecture
- PyTorch model development
- MIMIC-IV clinical time-series
- XGBoost, MLP, and Ridge baselines
- Hyperparameter optimization with Optuna
- Nested cross-validation
- Feature-selection experiments
- Ablation studies
- Regression evaluation with MAE, RMSE, R², and MedAE
An experimental reproduction of CoMAT-style mathematical reasoning using Qwen language models on the MMLU-Redux College Mathematics benchmark.
The project investigates model choice, generation parameters, reinforcement-learning-based fine-tuning, and reasoning-step importance.
- Qwen2 and Qwen3
- Mathematical reasoning with LLMs
- MMLU-Redux evaluation
- GRPO fine-tuning
- Prompt engineering
- Shapley-value-based reasoning-step analysis
- Experimental comparison of reasoning configurations
The strongest evaluated Qwen3 configuration achieved 69.70% accuracy on the experimental subset.
This project was developed collaboratively with Zeynep Naz Ödenir.
A data mining and machine learning study for predicting student dropout and academic success.
The project covers the complete workflow from exploratory analysis and feature selection to model tuning and ensemble learning.
- Exploratory Data Analysis
- Φk correlation-based feature analysis
- Random Forest
- XGBoost
- Gradient Boosting
- AdaBoost
- Support Vector Machines
- Neural Networks
- GridSearchCV
- Optuna
- Stacking ensembles
- 10-fold stratified cross-validation
- Feature-importance analysis
The project also examines class imbalance and the limitations of using overall accuracy for dropout prediction.
A Natural Language Generation project investigating few-shot table-to-text generation using Qwen3-4B and the ToTTo dataset.
Three prompting and structured-input strategies are compared using BLEU and BLEURT.
- Qwen3-4B
- ToTTo dataset
- Few-shot prompting
- Structured table representation
- Prompt-engineering experiments
- BLEU evaluation
- BLEURT evaluation
The best evaluated configuration achieved:
BLEU: 19.79
BLEURT: 0.551
This project was developed collaboratively with Zeynep Naz Ödenir.
A collection of five machine learning studies covering several major areas of modern machine learning.
Topics include:
- regression and model selection
- highly imbalanced classification
- dimensionality reduction
- representation learning
- clustering
- deep learning
- model interpretability
- adversarial robustness
The experiments use datasets including California Housing, Credit Card Fraud, MNIST, Wholesale Customers, Digits, and CIFAR-10.
Java projects focused on applying object-oriented programming concepts through domain-driven applications.
The repository includes projects involving:
- object-oriented design
- domain modeling
- encapsulation
- inheritance and polymorphism
- Java collections
- file I/O
- layered application structure
- separation of concerns
Python · Dart · Java · JavaScript
PyTorch · Scikit-learn · XGBoost · Optuna · Transformers · Hugging Face
Flutter · Next.js
Firebase · Firestore
Pandas · NumPy · Jupyter
Git · GitHub · BLE
- Software Engineering
- Machine Learning & Artificial Intelligence
- Natural Language Processing
- Mobile Application Development
- Healthcare Technology
- Intelligent Software Systems
- Connected Devices & IoT
- Data-Driven Applications
Additional academic work is available in:
These repositories contain selected coursework covering programming, machine learning, data mining, object-oriented programming, functional programming, and computational experiments.