E8IGHT
Jan 2024 – Present
Senior Software Engineer | Platform Development Team
Seoul, South Korea
Frontend platform and AI Native engineering
- Consolidated every product's frontend into one monorepo with shared config and dependencies, so new products start on domain work from day one
- Replaced the legacy CRA build with Vite and reworked the CI pipeline, cutting build time 80% (from 10 minutes to about 2)
- Stood up Sentry on internal servers to centralize frontend error monitoring and alerting across products
- Architected a design system on design tokens and shared components, and rebuilt the handoff from designers to code as an AI Native workflow
- Drove the team's shift to AI Native development, designing a frontend environment around subagents scoped by role, shared rules, and skills
- Introduced architecture guardrails and automated verification loops for coding agents, turning recurring failure patterns into enforceable rules
Hyundai Motor R&D data integration platform
Ontology knowledge graph platform for R&D data. Owned frontend architecture and development from PoC through production.
- Led frontend delivery of the core product: ontology data mapping, graph visualization, and the GraphRAG streaming UI
- Architected a graph visualization SDK shared across the company's GraphRAG products, with a core independent of any framework and a thin React adapter so each product plugs in its own data source
- Evaluated Cytoscape.js and Sigma.js for drawing the whole graph at once, judged that dense graphs hide the relations users need, and moved to an exploration UX that starts from key nodes and expands only what users ask for
- Kept d3-force for layout and wrote rendering and interaction directly on Canvas 2D, fitting nodes to the product and preserving layout on expansion so users keep their context
- Decoupled the graph simulation from React through an imperative API and a Zustand store, eliminating React renders during animation
- Owned release readiness in the client's secure network, covering service integrations, GitOps deployment on Harbor, Kubernetes, and ArgoCD, and SonarQube and Black Duck compliance
Digital Twin Federation integrated portal
Integrated portal for a national digital twin program under the Ministry of Science and ICT. Built over about two years as the sole full stack engineer on Next.js and NestJS.
- Chose a rendering strategy per page (SSG, ISR, dynamic SSR, RSC), so pages behind permissions are never cached and notices stay fresh through revalidation by tag
- Architected a Next.js BFF that holds tokens in HttpOnly cookies to prevent XSS token theft, with SameSite cookies and Origin checks guarding against CSRF
- Unified separate auth systems across subprojects into a single shell and embedded partner research institutes' tools through iframes, coordinating integration directly with each institute
- Designed RBAC with three roles and an organization model built around institutions, along with the admin console for users, notices, and boards
Company products (PMIS, NDX Cloud, and more)
PMIS
Solution for managing project schedules, resources, and drawings.
- Redesigned a tree of thousands of rows around lazy loading and virtual scrolling, cutting initial load from 12 seconds to 2 (about 80%)
- Moved drawing uploads from serial to parallel with optimistic UI and rollback on failure, cutting 19 uploads from 4 minutes to 20 seconds
NDX Cloud
Cloud platform for managing large image datasets.
- Abstracted four levels of permissions (organization, project, site, dataset) into a single RBAC matrix, unifying access control across every screen
- Optimized EXIF extraction to parse only the JPEG APP1 marker, cutting memory for 4,000 images from 20 GB to 800 MB
- Designed direct uploads from the browser to S3 with presigned URLs, halving upload time from 20 minutes to 10 and cutting server load about 70%