"Understand the input, trace the logic, verify the result."

Focusing on how systems behave under real-world conditions. I work on building predictable and reliable systems by identifying where assumptions fail and correcting the underlying logic.

Applied research on object detection for small and distant targets in aerial environments.

• Problem: Standard models failed in real scenarios due to a ~90% negative/background sample imbalance.
• Engineering Fix: Corrected data distribution and increased input resolution (640px → 1280px) to improve feature extraction for distant objects.
• Outcome: Established a unified benchmarking pipeline across YOLO v5–v11, enabling reproducible comparison of Precision / Recall / mAP / mAP@50 metrics for defense-oriented use cases.

Developing consistent tracking behavior for autonomous systems.

• Problem: Hardware-dependent frame rates caused unstable PID control behavior.
• Engineering Fix: Decoupled control logic from FPS using a delta_time (dt)-based update mechanism.
• Logic: Designed pixel-to-coordinate space mapping and intent-driven target selection, allowing full-scene perception while responding only to explicitly requested targets.

Using security research to understand internal system boundaries and protocol behavior.

• Program Analysis: Experience with Ghidra (static) and GDB (dynamic) for observing memory state and execution flow.
• System Auditing: Identifying logical weaknesses in state handling, authentication, and trust boundaries.
• Low-Level: Practical exposure to x86 assembly through reverse engineering.
• Approach: Focused on understanding system behavior rather than tool-driven exploitation.

• Comparative Study on Modern Object Detection Architectures
  A systematic analysis of performance trade-offs in YOLO-based models.
  Status: Study completed; currently awaiting official publication

I treat engineering as the process of making systems explainable.
If a system fails, I fix the underlying assumption, not just the symptom.