AI / ML
Engineer
M.Sc. in Artificial Intelligence – Bahçeşehir University (BAU) | AWS Certified Machine Learning Engineer – Associate (MLA-C01)
AI / ML Engineer and Senior Software Engineer with 11+ years of professional engineering experience, building production-ready AI systems across computer vision, LLMs, NLP, model serving, and AWS-based ML workflows. Strong hands-on experience in computer vision and document AI for risk-sensitive domains, including banking-domain document analysis, signature-region detection, signature verification, forgery-detection workflows, image preprocessing, segmentation-based cleanup, CNN / Siamese-network evaluation, synthetic data augmentation, threshold tuning, and false-positive / false-negative trade-off analysis.
Experienced in AWS-oriented ML engineering practices covering data preparation, feature engineering, model development, deployment orchestration, monitoring, and security-aware operations across SageMaker-related workflows, S3, Glue, Kinesis, Lambda, Step Functions, CloudWatch, Bedrock, Kendra, IAM, and KMS. Additional work includes LLM fine-tuning with PEFT / LoRA, security-oriented RAG systems, inference API design, model benchmarking, CI/CD integration, and production deployment using Python, PyTorch, TensorFlow, Hugging Face, FastAPI, Docker, OpenCV, Azure DevOps, and AWS-based ML services.
Education
M.Sc. Artificial Intelligence
Bahçeşehir University · 2021-2023
B.Sc. Computer Engineering
Toros University · 2010-2015
Core Skills
Generative AI & LLMs
LLM Fine-tuning (PEFT/LoRA, SFT), RAG, Prompt Engineering, BERT, Hugging Face Transformers, TRL, BitsAndBytes
Computer Vision
Custom CNNs, Siamese Networks, YOLO-based detection, U-Net segmentation (EfficientNet), ResNet, GANs & Diffusion Models, OCR
MLOps & Model Serving
FastAPI (REST/gRPC), Docker, Gradio, GitHub Actions, Azure DevOps, CI/CD Pipelines, Model Versioning & Evaluation, Prometheus, Grafana, Hugging Face Spaces
Cloud & Data Engineering
AWS (SageMaker Pipelines, Bedrock, Kendra, S3, Kinesis, Glue, Lambda, Data Wrangler), Apache Spark (PySpark)
Vector Databases & Storage
ChromaDB, Qdrant, SQLite, Amazon DynamoDB, MySQL, MongoDB, Fernet encryption, Feature Store management
Frameworks & Languages
PyTorch, TensorFlow, OpenCV, scikit-learn, Pandas, NumPy-Python, JavaScript/TypeScript, ReactJS, C#, PHP
Experience
Jun 2020 - Aug 2025
AI / ML Engineer | Senior Software Engineer
Halkbank
- Engineered an end-to-end banking computer-vision pipeline for document-to-signature verification, transforming heterogeneous PDF, TIFF, scanned, and image-based financial documents into high-fidelity image representations suitable for downstream ML models.
- Designed preprocessing workflows focused on preserving signature geometry, stroke structure, aspect ratio, resolution, and compression quality during document-to-image conversion, grayscale / black-and-white normalization, and region extraction.
- Fine-tuned YOLO-based object detection models to locate signature regions on complex banking documents containing handwriting, stamps, printed text, form lines, scan artifacts, and background noise; achieved above 99% internal validation performance for signature-region detection.
- Developed coordinate-based cropping logic using YOLO outputs to extract signature regions without meaningful image-quality loss, avoiding unnecessary resizing, distortion, or compression artifacts before verification.
- Fine-tuned a U-Net segmentation model to clean extracted signature crops by suppressing background contamination such as handwriting, stamps, printed text, document lines, scan noise, and visual clutter.
- Designed and evaluated CNN, Siamese-network, transfer-learning, and ensemble-based models for signature similarity analysis and forgery detection on real banking signature datasets.
- Explored GAN- and diffusion-based synthetic data augmentation strategies to improve model robustness under limited, imbalanced, and visually inconsistent genuine/forged signature data conditions.
- Standardized selected model components with ONNX/ONNX Runtime to reduce framework dependency friction and support a more consistent multi-model inference workflow across detection, segmentation, and verification stages.
- Built risk-sensitive evaluation workflows covering model benchmarking, threshold tuning, false-positive/false-negative trade-off analysis, and practical decision-boundary reasoning for banking fraud/forgery detection use cases.
- Contributed to production-minded ML delivery by combining document preprocessing, object detection, segmentation, synthetic augmentation, model comparison, inference-service integration, and operational reliability considerations across Python, PyTorch, TensorFlow, OpenCV, FastAPI, Streamlit Docker, and Azure DevOps.
Mar 2019 - Jan 2020
Software Engineer
Turkcell
- Architected modular cloud components and engineered seamless data integration layers between front-end interfaces and Java-based microservices.
- Optimized UI workflows across diverse multi-framework environments.
Aug 2017 - Mar 2019
Software Engineer
Orzed
- Built web applications and backend integrations using PHP and Python.
- Contributed to development and integration workflows for web- and mobile-based blockchain applications.
Aug 2015 - Aug 2017
Software Engineer
Blackhan
- Developed scalable web solutions prioritized for cross-platform compatibility and optimal system performance.
Selected Projects
All Projects →Security-First RAG Service for Swedish Tax Law
Apr 2026Designed and built a security-oriented RAG service using FastAPI, ChromaDB, SQLite, and Fernet-separating embeddings from encrypted source text through a split-storage architecture. Deployed with Gradio, Docker, GitHub Actions, Prometheus, Grafana, and Hugging Face Spaces.
Customer Support LLM-Baseline SFT with LoRA
Oct 2025Fine-tuned causal language models for enterprise English customer support using PEFT/LoRA and SFTTrainer with BitsAndBytes 4-bit/8-bit quantization. Developed high-performance batch inference utilities.
Human Segmentation-U-Net
Oct 2025Built a pixel-level human segmentation workflow using U-Net with an EfficientNet-B3 encoder for background removal and automated image-editing use cases.
Reinforcement Learning for Trading
Jan 2023Created trading agents trained with PPO and Actor-Critic methods in OpenAI Gym environments.
Certifications
AWS Certified Machine Learning Engineer-Associate (MLA-C01)
Mar 2026Google Cloud | Prompt Design in Vertex AI Skill Badge
Apr 2026Google Cloud Skills Badges-Responsible AI & LLM & Gen AI
Mar 2026AWS DynamoDB-AWS Training & Certification (Skillbuilder)
Apr 2026Model Monitoring Amazon CloudWatch-AWS Training
Jan 2026I am available for advisory engagements on AI architecture, LLM integration strategy, and production ML system design.