Experience

Work Experience

Waymo — Software Engineer Intern, Driver Understanding ML
May 2025 - Aug. 2025 | Mountain View, CA
  • Architected and deployed systematic prompt engineering and automation framework with Gemini LLM, serving millions of daily autonomous vehicle events, automating Traffic Events Triage with 91% recall on high-impact scenarios representing 35% of all queries.
  • Built production-grade few-shot inference pipeline, improving complex judgment recall by 9 points.
  • Established Gemini LLM prompt engineering and automation best practices adopted across 3 teams, enabling efficiency gain in automation tasks. Guides and practices were adopted by other colleagues, reporting 90% recall on a separate automation task.
  • Implemented dynamic prompting with context-aware sensory information, allowing context-aware automation.
US Army Research Laboratory (DEVCOM ARL) — Research Scientist Intern, Multimodal Foundation Models R&D
Jun. 2024 - Sep. 2024 | Adelphi Laboratory Center, MD
  • Led development of robust vehicle classification systems via sub-10M parameter foundation models trained on multimodal sensor data, achieving 12% improvement in detection accuracy while improving other downstream tasks performance (e.g. tracking) with real-time deployment on edge devices.
  • Led integration of physical decay models into foundation model pretraining, developing novel loss functions that improved robustness to environmental variations by 10%. Model was deployed and demonstrated on low-cost edge devices.
  • Deployed inference pipeline capable of real-time processing (sub-second inference latency) on edge devices, enabling distributed sensing applications for autonomous systems in resource-constrained environments.
Prometeia — Machine Learning Engineer
Jan. 2020 - Aug. 2022 | Istanbul, Turkey
  • Architected AI-based propensity scoring framework processing 3M+ customer transactions in a large bank, utilizing advanced time series embeddings and attention mechanisms to predict customer interests with improved accuracy.
  • Developed a deep learning credit default prediction system, engineering novel temporal transaction features that improved recall by 25%.
  • Enhanced Allianz Insurance's automated damage assessment system processing thousands of claims monthly, implementing state-of-the-art segmentation models with data augmentation pipeline that led to improved F1-score by 6%.
Turkish Aerospace — Software Design Engineer, Autopilot Systems Division
Jul. 2017 - Dec. 2019 | Ankara, Turkey
  • Developed and maintained signal processing libraries for autopilot control system software, reducing signal processing delay by up to 20%.
  • Led the interpretation of electromagnetic and vibrational noise within sensor data, developing signal filtering solutions compliant with control algorithms.
  • Created a sensor emulator framework, enabling realistic software-in-the-loop simulations for the autopilot department.

Research Experience

Graduate Researcher - Foundation Models
Aug. 2022 - Present | CyPhy Research Group, UIUC
Physics-Informed AI for Distributed IoT Systems
  • Developed foundation models integrating domain knowledge and physical laws with state-of-the-art self-supervised learning techniques for IoT Foundation Models.
  • Aiming to build efficient, lightweight, and explainable AI systems for IoT applications deployable on edge devices.
  • Collaborated with a multidisciplinary team, resulting in publications in top-tier conferences such as ACM SenSys, WWW, and NeurIPS.
Graduate Researcher - Adversarial ML & Security
Aug. 2020 - May 2022 | Systems Security Research Group, UIUC
Autonomous Vehicle & UAV Security Research
  • Designed embedded misbehavior detection framework (OVERTON) for V2X networks, integrating temporal anomaly detection, vehicular trust mechanisms, and novel ML architecture to combat adversarial attacks on autonomous vehicle communication.
  • Developed stealthy sensor-spoofing attack framework for UAV swarms, formulating ML-driven adversarial strategies that manipulate IMU readings to bypass control systems without triggering security alerts.
  • Expanded research on ML-driven adversarial mechanisms across vehicular and aerial platforms, leading to improved network security protocols and defensive strategies for autonomous systems.