I build practical AI systems that move from research notebooks to real production environments.
My work focuses on computer vision, multi-object tracking, segmentation, legal AI / RAG systems, remote-sensing workflows, and backend infrastructure for ML products. I enjoy designing end-to-end pipelines: from data ingestion, model experimentation, retrieval systems, and evaluation to FastAPI services, Docker/Kubernetes deployments, observability, and production iteration.
I am an AI/ML Engineer with experience across:
- Computer Vision: object detection, segmentation, multi-object tracking, real-time CCTV analytics
- RAG & NLP: hybrid retrieval, vector databases, legal AI systems, citation-aware generation
- Production ML: FastAPI services, async backends, Docker, Kubernetes/K3s, Redis, Celery
- EO/GIS Analytics: remote-sensing experiments with Sentinel/Landsat imagery and Google Earth Engine
- Research & Experimentation: medical imaging augmentation, NLP classification, geospatial AI workflows
I like building systems that are not only accurate, but also observable, maintainable, deployable, and usable.
| Computer Vision | YOLO / YOLOE, U-Net, GAN/DCGAN, OpenCV, BoT-SORT, ByteTrack, real-time video analytics |
| RAG & NLP | Hybrid retrieval, dense + sparse search, reranking, Qdrant, pgvector, Hugging Face, citation-aware LLM systems |
| Production Systems | FastAPI, PostgreSQL, Redis, Celery, Docker, Kubernetes/K3s, Linux, REST APIs |
| MLOps & Observability | OpenTelemetry, Prometheus, Grafana, structured logging, deployment pipelines |
| EO/GIS | Google Earth Engine, Sentinel/Landsat imagery, urban heat island analysis, geospatial segmentation |
Built production-oriented Turkish legal AI workflows that deliver cited, real-time answers over legislation and case law.
Key areas:
- Dual-pipeline RAG architecture for procedural law and case-law retrieval
- Hybrid retrieval with dense vectors and keyword search
- Cross-encoder reranking
- Numeric citation handling
- SSE-based LLM streaming
- FastAPI backend with PostgreSQL, pgvector, Redis, Celery, JWT/API keys, RBAC, rate limits, and audit logs
- Deployment on Kubernetes/K3s with TLS, monitoring, and observability
Designed and deployed a real-time vessel tracking system from overhead CCTV footage.
Key areas:
- Boat detection with YOLO-based models
- Travel-lift / boat-hoist detection with YOLOE segmentation
- Multi-object tracking with BoT-SORT
- Zone and line-crossing logic
- Debounce state machines for reliable counting
- Annotated output videos with live HUD, entry/exit counters, and event logic
Worked on Earth Observation and geospatial AI workflows for climate and urban analysis.
Key areas:
- U-Net segmentation workflows
- Sentinel/Landsat imagery processing
- Urban heat island analysis
- Google Earth Engine pipelines
- Reproducible training and evaluation workflows
- Full-stack interfaces for AI-assisted geospatial tools
Contributed to AI research projects involving medical imaging and NLP.
Key areas:
- DCGAN-based augmentation for limited-data medical imaging experiments
- Early HCC detection experiments
- NLP news classification pipeline
- Systematic feature/model iteration and experiment tracking
Worked on production legal AI and computer vision systems, including:
- Turkish legal AI platform with cited real-time answers
- RAG pipelines over legislation and case law
- Scraping, ingestion, embeddings, semantic chunking, and Qdrant indexing
- FastAPI backend architecture and production deployment
- Real-time marina vessel tracking using detection, segmentation, and tracking pipelines
Working on AI-assisted climate and urban analytics, including:
- EO/GIS segmentation workflows
- Remote-sensing data processing
- Google Earth Engine experimentation
- FastAPI + UI integration for tool-routing LLM workflows
Research work across:
- Medical imaging and data augmentation
- DCGAN-based synthetic data experiments
- NLP classification pipelines
- Reproducible experimentation and student mentoring
Worked on:
- KPI dashboards
- Automated reporting workflows
- Data analysis and business intelligence support
Worked on:
- Fraud-detection workflows
- Transformer-based NLP approaches
- REST API supported scoring workflows
Selected for Azure/AI-focused training and supported community learning and peer mentoring.
- Deep learning framework for urban seismic risk assessment: two-phase similarity algorithm for damage prediction and loss estimation — SPIE FST 2025, Oral Presentation
- From spectral indices to actionable insights: sensitivity analysis of a multispectral U-Net for spatially optimized urban heat island mitigation — SPIE FST 2025
- Harnessing EO and Natural Experiments for Urban Development: The UDENE Approach — GeoVISIONS 2025
- Developing a Virtual Laboratory for Climate Adaptation with the AI-Powered UDENE Tool — IAF GLOC 2026, Accepted
TOBB University of Economics and Technology
B.Sc. in Artificial Intelligence Engineering
Medium of instruction: 100% English
- Production-ready AI APIs
- Retrieval-augmented generation systems
- Computer vision pipelines for real-world camera setups
- Geospatial AI tools
- ML systems with clean architecture
- Developer-friendly AI products
- Research prototypes that can actually become usable software