Deep technical investigations bridging academic theoretical rigor with frontline engineering realities.
Google AdSense Slot (header)
Reserved Layout (CLS = 0) Skeleton Slot * Automatically active upon AdSense ID configuration
![[Research Briefing] Advancements in Artificial Intelligence Applications for Cardiovascular Disease Research](https://images.unsplash.com/photo-1667984390553-7f439e6ae401?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHw5fHxhcnRpZmljaWFsJTIwaW50ZWxsaWdlbmNlJTIwbWFjaGluZSUyMGxlYXJuaW5nJTIwY2xvdWQlMjBzZXJ2ZXJ8ZW58MHx8fHwxNzg5NDU3MzM1fDA&ixlib=rb-4.1.0&q=80&w=1080)
This report analyzes the achievements and limitations of AI applications based on CT, MRI, ECG, and ultrasound for cardiovascular disease diagnosis. It addresses the risks associated with the lack of input data validation and the necessity of model lightweighting for clinical deployment.
Jeonghyun ![[Research Briefing] Technical Debt Governance for Agentic AI Systems](https://images.unsplash.com/photo-1504639725590-34d0984388bd?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHw4fHxtYWNoaW5lJTIwbGVhcm5pbmclMjBzb2Z0d2FyZSUyMGNvZGUlMjBhbGdvcml0aG0lMjBwcm9ncmFtbWluZ3xlbnwwfHx8fDE3ODk4MjIxMTd8MA&ixlib=rb-4.1.0&q=80&w=1080)
This briefing establishes the concept of 'Stochastic Tax'—the recurring operational cost—and cumulative design debt caused by the stochastic nature of Agentic AI. It proposes a reference architecture and a graduated autonomy governance strategy to manage these risks.
Jeonghyun ![[Research Briefing] The Evolution of Multimodal AI Evaluation: From Recognition to Reasoning](https://images.unsplash.com/photo-1667984390527-850f63192709?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHwxM3x8YXJ0aWZpY2lhbCUyMGludGVsbGlnZW5jZSUyMG1hY2hpbmUlMjBsZWFybmluZyUyMGNsb3VkJTIwc2VydmVyfGVufDB8fHx8MTc4OTczNjk5NHww&ixlib=rb-4.1.0&q=80&w=1080)
This analysis examines the evolution of multimodal AI evaluation frameworks through a four-stage framework, transitioning from simple object recognition to complex cognitive reasoning. It highlights the limitations of static evaluation and proposes a paradigm shift toward living benchmarks and Embodied AI evaluation.
Jeonghyun ![[Research Briefing] AgentOps: A Framework for Controlling Uncertainty in Agentic AI Systems](https://images.unsplash.com/photo-1739036868260-c26b292cd85d?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHwxMXx8YXJ0aWZpY2lhbCUyMGludGVsbGlnZW5jZSUyMGFsZ29yaXRobSUyMGRhdGElMjBjb21wdXRlfGVufDB8fHx8MTc4OTUwMTIwOHww&ixlib=rb-4.1.0&q=80&w=1080)
Proposes the AgentOps framework and a six-stage automation pipeline to manage the non-deterministic behavior of agentic AI. It addresses strategies for controlling system uncertainty by integrating perspectives from developers to business users.
Jeonghyun ![[Research Briefing] SHIELD: An Efficient ML Framework for Healthcare IoT Anomaly Detection](https://images.unsplash.com/photo-1709120395858-92f1c7c577f5?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHw3fHxhcnRpZmljaWFsJTIwaW50ZWxsaWdlbmNlJTIwbWFjaGluZSUyMGxlYXJuaW5nJTIwY2xvdWQlMjBzZXJ2ZXJ8ZW58MHx8fHwxNzg5NDU3MzM1fDA&ixlib=rb-4.1.0&q=80&w=1080)
This research analyzes eight ML models through the SHIELD framework to secure healthcare IoT. It demonstrates that XGBoost and KNN are the optimal choices in terms of detection performance and computational efficiency.
Jeonghyun ![[Research Briefing] Addressing Concept Drift via XAI-based Profile Drift Detection (PDD)](https://images.unsplash.com/photo-1618401471353-b98afee0b2eb?w=1200&auto=format&fit=crop&q=80)
This article presents the PDD methodology, which utilizes Partial Dependence Plots (PDP) to detect concept drift. It discusses strategies to enhance MLOps stability by visualizing and quantifying changes in variable relationships, overcoming the limitations of performance-metric-based detection.
Jeonghyun ![[Research Briefing] Adaptive MLaaS Composition Framework for IoT Environments](https://images.unsplash.com/photo-1591696331111-ef9586a5b17a?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHw2fHxhcnRpZmljaWFsJTIwaW50ZWxsaWdlbmNlJTIwbWFjaGluZSUyMGxlYXJuaW5nJTIwY2xvdWQlMjBzZXJ2ZXJ8ZW58MHx8fHwxNzg5NDU3MzM1fDA&ixlib=rb-4.1.0&q=80&w=1080)
This research analyzes an adaptive composition framework that restores the performance of MLaaS combinations—which degrade due to concept drift and evolving system requirements in IoT environments—through incremental replacement rather than total reconfiguration. It highlights experimental results achieving over 95% accuracy across MNIST, FMNIST, and HAR datasets using a Service Assessment Model (SAM) and CMAB algorithm, while noting the limitation of relying on a 1:1 replacement strategy.
Jeonghyun ![[Research Briefing] Multimodal ML Analysis Framework for GaSe MBE Growth](https://images.unsplash.com/photo-1667984390538-3dea7a3fe33d?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHw1fHxhcnRpZmljaWFsJTIwaW50ZWxsaWdlbmNlJTIwbWFjaGluZSUyMGxlYXJuaW5nJTIwY2xvdWQlMjBzZXJ2ZXJ8ZW58MHx8fHwxNzg5NDU3MzM1fDA&ixlib=rb-4.1.0&q=80&w=1080)
Introducing an ML framework that optimizes the GaSe thin-film growth process by combining real-time RHEED diagnostics with XRD/AFM analysis. It presents the possibility of real-time process control by analyzing the data-driven impact of growth variables on thin-film quality.
Jeonghyun ![[Research Briefing] SecMLOps: Integrated Security Framework for the Machine Learning Lifecycle](https://images.unsplash.com/photo-1777785113328-1ec3664f5516?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5OTUzNTN8MHwxfHNlYXJjaHwxfHxkYXRhJTIwY2VudGVyJTIwc2VydmVyJTIwbWFjaGluZSUyMGxlYXJuaW5nJTIwY2xvdWQlMjBpbmZyYXN0cnVjdHVyZXxlbnwwfHx8fDE3ODk0OTEzNzR8MA&ixlib=rb-4.1.0&q=80&w=1080)
We propose the SecMLOps framework to integrate security across the entire MLOps pipeline. Through a pedestrian detection system case study, we analyze the tradeoff between security and performance and present optimization strategies.
Jeonghyun 
A deep technical dissection of the landmark 2024 RAND Corporation research report. Unpacking the 5 structural failure modes and the fundamental communication gap between engineering and the C-suite.
Jeonghyun 
Dissecting the exact technical mechanisms behind silent model failure in high-throughput enterprise serving, and architecting resilient real-time drift mitigation pipelines.
Research Fellow: MLOps Lab 
Why traditional deterministic Agile roadmaps fail when applied to probabilistic machine learning systems, and how AI Product Leaders navigate PoC purgatory.
AI Product Strategy Practice