Applied Machine Learning Engineer · Computer Vision · Python APIs
I build reproducible models, reliable APIs, and user-facing AI products.
I am a Management Information Systems graduate based in Samsun, Türkiye. My main focus is taking applied AI projects through the complete engineering lifecycle: dataset auditing, leakage-safe splitting, model training, calibrated evaluation, API delivery, visual interfaces, and reproducible documentation.
- Computer vision: classification, object detection, medical imaging, edge AI
- ML engineering: PyTorch, YOLO, ONNX/LiteRT, calibration, experiment design
- Product delivery: FastAPI, React/Next.js, Flutter, Docker, CI and CodeQL
- Data systems: Python ETL, PostgreSQL, Power BI, validation and analytics
I am open to entry-level and junior opportunities in applied machine learning, computer vision, Python backend, and data-product engineering.
| Project | Verifiable result | Product surface |
|---|---|---|
| FractureLens | DenseNet121 test AUROC 0.912; calibrated classification + localization | FastAPI and browser demo |
| HemaScope | Segmentation mean Dice 0.937; calibrated ECE 0.0033 | Mask, morphology and selective-prediction UI |
| DentXplain | C1 final-cohort F1 0.463; conditional FDI accuracy 0.957 | Annotated panoramic X-ray review UI |
| SmartWaste Vision | 90.80% test accuracy; PyTorch → ONNX → LiteRT comparison | Offline Flutter application |
| KentLens | Reproducible urban ETL/API pipeline; 18/18 tests passing | Power BI and FastAPI |
Medical-imaging projects are research prototypes and are not intended for diagnosis or treatment decisions.
|
A duplicate-aware fracture system combining calibrated classification, local detection, fusion logic, error analysis, and a visual demo.
|
A calibrated white-blood-cell system combining multi-class segmentation, measurable morphology, selective classification, and an input-support guard.
|
|
An anatomy-constrained panoramic dental X-ray pipeline for pathology detection, FDI numbering, confidence tiers, and auditable review.
|
An offline waste classifier with measured accuracy, latency, and model-size trade-offs across PyTorch, ONNX Runtime, LiteRT, and Flutter.
|
|
A local-first DevSecOps orchestrator that converts real scanner output into deterministic policy decisions, SARIF, and privacy-preserving reports.
|
An urban analytics platform joining tested Python ETL, PostgreSQL, FastAPI, data-quality controls, and decision-oriented Power BI reporting.
|
- Freeze data splits and thresholds before final evaluation.
- Report limitations, calibration, error cases, and negative results.
- Keep datasets, credentials, and restricted artifacts outside Git.
- Ship tested interfaces instead of stopping at a notebook.
- Automate quality checks with CI, dependency monitoring, and CodeQL.
Explore all repositories →
Turning research ideas into measurable, documented, and usable software.