A short introduction and the purpose of this portfolio site.

Retrieval-Augmented Generation system answering questions about personal ML and automation projects — combining AST-extracted code structure with hand-curated documentation in a hybrid knowledge base, featuring similarity-threshold gating and project-filtering metadata routing.

End-to-end big data pipeline for classifying factory operator actions from 10 GB, 200 Hz motion-capture sensor streams — covering out-of-core engineering, feature extraction, supervised/unsupervised learning, and real-time streaming with concept drift detection.

Design and implementation of a desktop RPA suite using Selenium and Tkinter, featuring a local mock ERP test-harness for offline end-to-end testing, dynamic retry engines, and multi-threaded GUI systems used to power collectibles micro-venture sync workflows.

A platform-agnostic, config-driven mobile automation suite built with Python and Appium, designed to manage over 1,000 active listings across Turkish C2C marketplaces using coordinate-based pairing and overlap detection algorithms.

End-to-end white blood cell classification using DenseNet121 with custom WBCAttention and MedSwish layers, achieving 98.53% in-distribution and 89.05% OOD accuracy via retraining-free inference adaptation, featuring a 5-step Medical Enhanced Filter (MEF) pipeline and an autonomous LLM shortcut-detection agent.

Real projects in ML, backend, and automation with measurable outcomes.