A — Statement

Physicist by training. Engineer by craft. I build the plumbing that lets AI systems see, act, and learn at scale.

I turn messy web workflows into programmable APIs — and pixels into decisions machines can trust.

GitHub

Resume

B — Background

AI Engineer with 2+ years of hands-on experience building production ML systems, scalable data infrastructure, and browser automation platforms. Architected distributed data collection networks on Kubernetes, developed real-time Detect & Avoid systems for autonomous drones, and engineered CDP-based skill systems turning web workflows into programmable APIs for AI agents. Physics background provides strong foundations in mathematical modeling and first-principles problem solving.

C — Work

◆Architected distributed data collection infrastructure with Docker containers and rotating proxy networks, processing millions of records across multiple cloud nodes on GCP.
◆Engineered Chrome DevTools Protocol (CDP) skill system converting complex web workflows into programmable CLI APIs for AI agents — 100+ commands across 30+ skills powering a Claude Code plugin.
◆Built Android CDP skill: ADB-based CLI giving AI agents control over mobile devices — UI automation, app management, HTTPS interception via mitmproxy, and SSL unpinning with Frida.
◆Developed anti-detection solutions including TLS fingerprint impersonation, session management, and browser-routed API requests to handle WAF/CDN protections at scale.
◆Built investor probability models and founder classifiers using vector similarity search and decision tree ensembles for deal intelligence.

◆Developed real-time Detect and Avoid computer vision system for autonomous drones using PyTorch and OpenCV, enabling safe flight in complex airspace environments.

◆Built image preprocessing pipelines and trained segmentation models (U-Net variants) and classical ML classifiers, directly improving product accuracy and user experience.
◆Designed and implemented a weak labeling system that reduced manual annotation effort by ~60%, accelerating model iteration cycles from weeks to days.
◆Deployed models via CI/CD pipeline, managing the full lifecycle from training to production serving.
◆Documented APIs and codebases with Sphinx, cutting new developer onboarding time.

◆Implemented ResNet and VAE architectures in PyTorch for denoising InSAR satellite imagery, improving signal-to-noise ratio for earthquake deformation analysis.
◆Applied signal processing techniques (Wavelet Transforms, STA/LTA, Template Matching) to seismological datasets, combining physics domain knowledge with ML methods.

D — Stack

I reach first for Python and PyTorch for modeling, OpenCV and scikit-learn when a problem is small enough to stay classical, and TensorFlow when a project demands it. Vision work — object detection, segmentation, signal processing — is where I've spent the most time.

Around the models: Docker / Kubernetes on GCP and Cloud Run, FastAPI services, CI/CD for the MLOps layer, and Chrome DevTools Protocol + Playwright for browser automation. Also fluent in TypeScript, C++, Go, and SQL.

E — Education

Sep 2019 — Jun 2026

Boğaziçi University

B.Sc. Physics

Machine Learning Applications in Physics (Symbolic Regression, GNNs, RBMs), Machine Learning (SVM, CNN, NNs), Numerical Methods, Computer Vision, Computer Networking, Group Theory

Sep 2017 — Jun 2019

Kadir Has University

Vocational Degree, Computer Programming

Operating Systems, Object-Oriented Programming (Java), Database Systems (SQL)

Certifications

— Quantum Optics 1 & 2 (École Polytechnique)
— Qiskit Global Summer School 2022
— Neural Networks and Deep Learning
— Structuring ML Projects