Dinko Dermendzhiev

Under the guidance of Professor Brendan D. Saltaformaggio, PhD.

• Developed a novel forensics technique to recover Deep Learning models from volatile memory (Python, Volatility, PyTorch, C, LiME).
• Evaluated on 22 SOTA DL models from language and vision domains, achieving 100% recovery accuracy and model reuse (e.g. Llama 3, YOLOv10).
• Designed a log-enhanced symbolic execution technique for reconstruction of Linux program executions (angr, IDA Pro, nginx).
• Engaged in research paper analysis, ideation, project design, collaboration, and presentation during weekly lab meetings.

Publication:
Achieving Zen: Combining Mathematical and Programmatic Deep Learning Model Representations for Attribution and Reuse.
D. Oygenblik, D. Dermendzhiev, F. Sofias, M. Yao, H. Xu, R. Zhang, J. Park, A. K. Sikder, and B. Saltaformaggio.
In Proc. 2026 Annual Network and Distributed System Security Symposium (NDSS), San Diego, CA, 2026.

• Engineered a context-aware, feedback-driven, agentic system for automated firmware emulation debugging and refinement (LLMs, RAG, Agentic AI).
• Automatically applied interventions to refine several embedded Linux firmware in record time to a successful web server startup stage.
• Enhanced reliability and feature set of IoT embedded firmware rehosting framework and HPC infrastructure for testing at scale (Python, bash).

Publication:
Target-Centric Firmware Rehosting with Penguin.
A. Fasano, Z. Estrada, L. Craig, B. Levy, J. McLeod, J. Becker, C. Kline, E. Witham, C. DiLorenzo, A. Bobi, D. Dermendzhiev, T. Leek, W. Robertson.
Workshop on Binary Analysis Research (BAR) 2025, San Diego, CA, 2025.

• Supported Computing Systems and Digital Design Lab courses, focusing on Assembly, C programming, VHDL, and FPGA design.
• Guided young engineers through problem solving, troubleshooting, and testing of SW/HW components and tools.
• Administered an ECE Technical Communications Protocol (collaboration, documentation, presentation skills) to 50+ students every semester.

• Researched and implemented secure communications for Car and Key fob firmware for the MITRE embedded Capture the Flag competition.
• Reverse engineered ARM 32-bit binaries on Red Team with Ghidra, employing penetration testing techniques.

Publication:
2023 MITRE eCTF Design Documentation and Attack Approach.
D. Dermendzhiev, K. Paton-Smith, A. Devakonda, L. Doyle, L. Estrella, V. Gonugondla, R. Verma, J. Zhang.
2023 MITRE Embedded Capture the Flag, 2023.

• Developed 2 ML solutions to automate server health monitoring, decreasing incident response time by 90% (T-SQL, C#, .NET, XGBoost).
• Automated ML model use and maintenance (data retrieval, pre-processing, inference, batching, finetuning).
• Designed and simulated a database migration from SQLServer (relational) to CosmosDB (JSON) optimizing for operational use.
• Built a CRUD web app to evaluate the real-time performance of common database operations, showing 75% decrease in response times.