I am a Computer Engineering student and undergraduate AI researcher from Türkiye, focused on building robust machine learning systems for computer vision, medical image analysis, and domain generalization.
I work on research-driven AI systems that combine strong experimental methodology with practical deployment. My interests include shortcut learning, self-supervised learning, semi-supervised learning, model calibration, and domain-independent classification models for chest radiographs and heterogeneous medical datasets.
I enjoy working across the full AI lifecycle: dataset construction, model training, evaluation, calibration, optimization, deployment, and reproducible research pipelines.
- Medical image analysis with chest radiographs and DICOM/PACS workflows
- Domain generalization and shortcut learning in healthcare AI
- Fine-tuning open-source models for task-specific use cases
- Preparing deployment-friendly model variants with ONNX, quantization, and MLX-compatible workflows for Apple Silicon
- Vision Transformers, ConvNeXt, Swin, DINO-based models, and calibration methods
- Offline reinforcement learning for clinical decision-making research
- Building reproducible AI demos with FastAPI, Gradio, Hugging Face Spaces, and cloud deployment
| Project | Description | Stack |
|---|---|---|
| AutoLens AI | Vehicle body-type classification system comparing CNN baselines and DINOv3 ViT-S/16 variants. The final calibrated model was deployed with ONNX and Hugging Face Spaces for reproducible inference. | Python, PyTorch, DINOv3, ONNX, FastAPI, Gradio |
| MIMIC Sepsis Offline RL | Research-style offline reinforcement learning study on MIMIC-IV v3.1 Sepsis-3 ICU data, modeling treatment as a 62-feature state and 25-action MDP evaluated with FQE/WIS. | Python, PyTorch, Offline RL, CQL, MIMIC-IV |
| MetroCast AI | End-to-end AI-powered weather forecasting system with deep learning training, ONNX Runtime inference, Rust Axum backend, AWS deployment, and React dashboard. | Python, Rust, ONNX Runtime, AWS, React |
| Apex | Local-first multi-agent financial research cockpit combining RAG, ensemble ML, signal analysis, and market narrative summarization. | Python, LangGraph, RAG, Ensemble ML |
| CodeContext | Developer tool that converts codebases into structured LLM-ready context with browser-side processing and GitHub repository support. | TypeScript, Next.js, React, Tailwind CSS |
Core: PyTorch, TensorFlow, Scikit-learn, Pandas, NumPy, OpenCV, ONNX, ONNX Runtime, WandB Research Areas: Computer Vision, Medical AI, Domain Generalization, Offline RL, RAG, LLM Workflows
Backend & Web: FastAPI, Flask, Axum, React, Tailwind CSS Tools: Git, Docker, Linux, AWS EC2/S3/IAM, LaTeX, Hugging Face Spaces, Gradio
- Robust medical AI under dataset shift
- Domain generalization for chest radiograph classification
- Vision Transformers and self-supervised visual representations
- Model calibration and uncertainty-aware evaluation
- Offline reinforcement learning for retrospective clinical research
- Local-first AI agents, RAG systems, and reproducible ML pipelines
- Task-specific model fine-tuning and local inference optimization
Building research-driven AI systems from data to deployment.