Research
I am a full-time researcher at the Software Convergence Research Center (SCRC), Kookmin University, under the supervision of Professor Park Soo-hyun. My research addresses the challenge of deploying intelligent AI systems in resource-constrained and connectivity-challenged environments — particularly in special communication networks such as underwater acoustic and polar satellite relay systems.
Research Areas
Special Communication Networks
Conventional communication infrastructure assumes stable, high-bandwidth links. Many critical environments — underwater acoustic channels, polar satellite relay networks, and deep-field sensor arrays — operate under severe latency, bandwidth, and reliability constraints. I study the system-level requirements for embedding AI capabilities into these environments.
Current work focuses on characterizing the inference and data-handling demands of AI tasks under channel constraints, with the aim of defining deployable model and runtime specifications for such networks.
Edge AI & Lightweight Language Models
Large language models are computationally prohibitive for deployment on edge devices. I investigate strategies for adapting small language models (SLMs) and quantized LLMs to operate within the memory, power, and compute budgets of resource-constrained hardware — from embedded processors to communication relay nodes.
This includes evaluating quantization techniques (GGUF, GPTQ, AWQ), studying the quality-size trade-off in distilled models, and assessing practical feasibility of on-device inference in communication-oriented use cases.
System-Level Implementation & Optimization
Research prototypes must be grounded in real implementations. I build and maintain inference runtimes and evaluation tools that make performance constraints concrete: latency measurements, memory footprint, and cross-platform reproducibility.
Active projects include llmrc (a Rust-based local LLM runtime with C++ FFI and GGUF model loading) and ml-engine (a C++/LibTorch inference pipeline with a lightweight REST interface).
Physical AI Applications
Physical AI refers to the integration of AI inference into physical systems and infrastructure where real-time reliability and resource efficiency are hard constraints. I am interested in defining application models and performance requirements for AI in communication infrastructure: autonomous relay management, signal classification, and anomaly detection in hardware-constrained network nodes.
Selected Publications
Research in progress — no publications finalized at this time.
Full academic record on CV.
Collaboration
I am open to research collaborations with groups working on special communication systems, edge computing, or AI for resource-constrained environments. I am also interested in connecting with researchers at the intersection of communication networks and machine learning systems. Please reach out via the contact page.