CV

Jihwan Park

Systems engineer interested in reliable infrastructure for constrained environments. My work connects restricted-network platform migration, stateful service recovery, private artifact supply chains, operational AI serving, and resource-constrained AI systems.

I focus on systems infrastructure that enables internal services and AI workloads to operate reliably under network, storage, hardware, and deployment constraints, rather than on AI model architecture itself.

GitHub

Interests

Research and Engineering Interests

The main direction is reliable systems and infrastructure for services that must run under real operational constraints.

Reliable Systems / Platform Engineering

Service reliability, deployment repeatability, recovery planning, and cloud-native operations for constrained internal environments.

Restricted-Network Infrastructure

Private artifact workflows, image distribution, repository behavior, bootstrap dependency paths, and deployment automation in restricted networks.

Storage / Stateful Recovery

Stateful workload recovery, storage attach/detach behavior, RTO/RPO-aware service design, and the practical boundary between hot failover and recoverable systems.

Resource-Constrained AI Systems

TinyML memory management, hardware-aware CNN acceleration, full-GPU workload constraints, degraded serving, and infrastructure around AI workloads.

Education

Education

A software and hardware background for systems work across the stack.

Kyung Hee University

Bachelor's degree in Computer Science and Engineering, with a double major in Electronic Engineering.

GPA: 4.01 / 4.3 ยท Top 5% graduate

Academic Excellence Award

2021 - 2025

Experience

Infrastructure Experience

Publicly described at the level of constraints, architecture choices, and technical direction. Sensitive operational details are intentionally generalized.

Restricted-Network Platform Migration

2025 - Present

  • Migrated selected services from ad-hoc Docker-based operation toward a lightweight Kubernetes/K3S-based internal platform under restricted network conditions.
  • Worked through deployment standardization, internal image distribution, service exposure, persistent storage, and operational recovery concerns.
  • Focused on practical reliability in an environment where external registries, public package repositories, public DNS, and managed cloud services were not freely available.

Stateful Service Recovery Analysis

2025 - Present

  • Analyzed the gap between zero-downtime hot failover and fault-tolerant recovery for services that depend on persistent storage.
  • Reasoned about Longhorn-backed PVC behavior, including storage detach/attach delay, recovery time, service downtime, and RTO/RPO-aware operating models.
  • Used the experience to distinguish stateless rescheduling from storage-aware recovery planning for stateful internal services.

Private Artifact and Bootstrap Supply Chain

2025 - Present

  • Worked through private container registry, Nexus, image mirroring, and internal artifact distribution for restricted-network infrastructure operation.
  • Considered bootstrap dependency risks such as registry or Nexus self-dependency, where a failed artifact service must not depend only on itself for recovery.
  • Investigated Maven/Nexus repository behavior, including how uploaded artifact paths can be governed by POM GAV metadata rather than only the chosen upload path.

Internal Delivery and Operational Automation

2025 - Present

  • Evaluated internal CI/CD and GitOps-style delivery flows connecting source changes, image builds, registry pushes, manifest updates, synchronization, and rollback.
  • Used Git history and declarative configuration as a way to reason about deployment traceability, reproducibility, and recovery from failed changes.
  • Connected service routing, storage, registry, artifact repositories, and deployment automation into a more coherent operating model.

Full-GPU Serving Availability Strategy

2025 - Present

  • Evaluated AI serving operation where fine-grained GPU sharing was not always practical because production workloads required full-GPU allocation.
  • Reasoned about degraded serving as a realistic availability model: accepting lower capacity during failure while keeping the service alive.
  • Compared the practical roles of full-GPU placement, active-active service replicas, MIG, time-slicing, and load-balancer-level traffic rerouting.

Projects

Resource-Constrained Systems Projects

Graduation-level projects that connect lower-level resource constraints with AI workload execution.

Lifecycle-Aware Memory Analysis for TensorFlow Lite Micro

TinyML runtime memory management project that analyzed tensor lifetimes, conflict relationships, BFS-style allocation ordering, and in-place update opportunities. Project evaluation reduced peak memory by 28.33% on Person Detect and 14.06% on Tiny U-Net compared with the original size-based allocation plan.

Area-Efficient CNN Accelerator Using Bit-Shift Operations on Zynq FPGA

Hardware-aware AI inference project using bit-shift operations, INQ/STE-based quantization, IM2COL, systolic array dataflow, and AXI4-Lite PS/PL integration. Project evaluation reported 36.5% FPGA area decrease and 8.83ms PS+PL runtime compared with 15.9ms PS-only execution.

Skills

Technical Skills

Grouped by the systems layers that show up repeatedly in the work.

Systems / Infrastructure

Linux, Docker, Kubernetes/K3S, Longhorn, private registry, Nexus, internal CI/CD, GitOps concepts, Traefik, NodePort/hostPort, TLS/HTTPS.

Data / Backend Systems

MinIO, ClickHouse, MongoDB, RabbitMQ, asynchronous processing, SQL.

Programming

C/C++, Python, JavaScript, TypeScript, Java, Go, SQL, Verilog.

Hardware / Embedded

FPGA, Zynq, Vivado, Quartus, STM32, Arduino, Raspberry Pi.