About

I'm a Software Engineer at SK hynix America, building secure internal AI applications with Python, local LLMs, and cloud AI models.

I develop tools for document analysis, market and competitive intelligence, and meeting summaries while protecting sensitive proprietary company information. I also improve model performance through AI evals, refined prompting, context definition, and other techniques to ensure factual accuracy and eliminate hallucinations.

I have experience developing MES software for large‑scale EV battery production. I hold a B.S. in Computer Science from UC San Diego.

Education

  • B.S. Computer Science, 2025

    UC San Diego, Jacobs School of Engineering

  • Santa Monica College, 2021

    Overall GPA: 4.0

Skills

  • AI & LLM: Qwen, Llama, EXAONE, Gemini API, Ollama, vLLM, RAG architectures, Familiarity with LoRA
  • Languages & Frameworks: Python, C#, SQL, FastAPI, .NET, WinForms, Java, C++, HTML, CSS
  • Databases & Testing: Oracle Database, EMQuest, 5G FR1/FR2 OTA Testing, Regulatory Testing

Experience

SK hynix America · San Jose, CA

Software Engineer, AI Applications

May 2026 — Present

Develop secure internal AI applications using Python, local LLMs (Qwen, Llama, EXAONE, etc.) and cloud models (Gemini API) while protecting sensitive proprietary company information.

  • Experience with AI‑serving frameworks (Ollama, vLLM).
  • Working knowledge of RAG architectures (chunking, retrieval, reranking, evaluation metrics).
  • Familiarity with model fine‑tuning (LoRA).
  • Built Report AI to analyze multi‑team weekly reports and generate concise summaries for the executive team.
  • Developed Market Intelligence AI to track market changes, customers, competitors and industry developments, and their impact on SK hynix.
  • Built Meeting AI to capture key discussions, decisions, action items, and follow‑up tasks from business meetings.
  • Improved model performance through AI evals, refined prompting, context definition, fact extraction, citation validation and grounding to ensure factual accuracy and eliminate hallucinations.
  • Packaged and deployed applications through AI Hub, providing one‑click installation, automated updates, and secure delivery to company laptops.

SK AX USA, Inc. · Glendale, KY

MES Software Engineer

Jul 2025 — Apr 2026
  • Developed MES (Manufacturing Execution System) software features and enhancements to support large‑scale electric vehicle battery production at SK Battery America and BlueOval SK.
  • Ensured seamless integration with other manufacturing applications, such as WMS and ERP.
  • Ensured stable 24/7 operations by debugging and troubleshooting manufacturing applications and issues.
  • Collaborated with local site engineers, equipment teams, and internal and external IT partners to deliver customized MES solutions for business partner needs.
  • Promoted from Associate Engineer position in Georgia.

SGS North America · San Diego, CA

OTA Engineer, Part‑Time

Jun 2024 — Jun 2025
  • Conducted Radio Frequency Over‑the‑Air (OTA) antenna testing for major clients including Google and Samsung on mobile products (CDMA, GSM, LTE, Wi‑Fi, 5G), ensuring compliance with client and SGS quality system requirements.
  • Performed debugging, test engineering, and system management for wireless devices using EMQUEST.
  • Maintained industry standard test platforms, resolved issues, and ensured compliance with accreditation.
  • Collaborated with clients and test equipment manufacturers to resolve testing issues and deliver high‑quality service.

Alpha Gamma Sigma, Santa Monica College

Web Developer

Sep 2019 — Jun 2021
  • Developed and maintained a student database to manage member information, volunteer records, and membership fees.
  • Built and managed the official website and e‑commerce platform, including an event calendar and digital newsletter library.

Republic of Korea Army

Administrative Support Specialist, Sergeant

Dec 2014 — Sep 2016

Managed administrative tasks and records using military databases and software, ensuring data security and integrity.

Kmong Inc.

Technical Consultant, Self‑employed

Jan 2013 — Dec 2014

Advised startups on web strategies and platform development to drive business success.

Gaming / Anime Community Website

Founder

Oct 2010 — Dec 2014

Led operations to grow this website to 100k+ users and 40M+ visits in three years. Organized anime expo events.

Projects

Internal AI applications built at SK hynix America.

Report AI icon

Python · LLM Pipeline · Automation

Report AI

Analyzes multi‑team weekly reports and generates concise summaries for the executive team.

Highlights
  • Eliminated 2–4 hours of manual copy‑paste per reporting cycle by automating a 5‑tier Korean Word‑document hierarchy (team member → team lead → department → director → CEO) into a single compiled, template‑styled .docx.
  • Designed a citation‑anchored generation pipeline (PREP → GENERATE → CHECK → VERIFY → REPAIR) where every generated bullet must cite the numbered source lines it used; deterministic regex checks catch hallucinated and dropped facts with zero LLM cost, and a small‑model verifier judges each bullet against only its own cited lines.
  • Cut per‑topic generation from 220s to ~40s (5.5×) by profiling the real bottleneck — the small model's discarded reasoning tokens and model swapping, not the 14B generator — then switching to a non‑thinking instruct model and pinning both models in memory.
  • Built an offline eval harness that scores fact coverage, hallucination rate, and style compliance against real documents, so model and prompt changes are decided with numbers rather than spot‑checks.
  • Handled DRM/IRM‑encrypted documents by falling back to Word COM automation using the user's own credentials when standard parsing fails.
Market Intelligence AI icon

Gemini · GitHub Actions · Flask

Market Intelligence AI

Tracks market changes, customers, competitors and industry developments, and their impact on SK hynix.

Highlights
  • Delivered a daily long‑form intelligence report on customers, competitors, and the memory/AI market for leadership, generated from grounded web search across tiered sources plus a Telegram feed of industry channels.
  • Solved a hard network constraint architecturally: the company laptop can read GitHub over HTTPS but cannot make outbound POSTs. Moved every internet‑touching operation into GitHub Actions, so the laptop only ever pulls results — the workaround became the architecture.
  • Extracted Gemini's grounding metadata into per‑claim citation links so every statement in the report is traceable to a real source, deliberately matching on text snippets rather than the API's byte offsets (which silently corrupt on Korean multi‑byte text).
  • Built a local Flask viewer with per‑company Radar pages, an executive top‑3 Signals layer, archive search, and SK hynix house‑template .docx export.
Meeting AI icon

Whisper · pyannote · Playwright

Meeting AI

Captures key discussions, decisions, action items, and follow‑up tasks from business meetings.

Highlights
  • Built a fully local pipeline: a Playwright bot joins the Webex call as a self‑announcing silent participant, captures system audio via WASAPI loopback, and post‑call runs local Whisper (large‑v3) transcription concurrently with pyannote speaker diarization, merged by timestamp.
  • Redesigned extraction into three passes after testing showed schema‑constrained generation collapses into empty or placeholder output regardless of model size: free‑form synthesis, then mechanical structuring with line‑number citations, then a grounding check that drops any claim whose cited transcript line doesn't exist or doesn't support it — including fabricated attendees.
  • Shipped real‑time transcription during the meeting using a small Whisper model over a sliding window with LocalAgreement‑2 confirmation, streamed over SSE, plus a guard that discards Whisper's decoding‑loop hallucinations before they can be confirmed permanently.
  • Rebuilt the product as a notes‑first app (type your own notes; the AI summary lands beneath them), using the filesystem as the database with a global write lock and content‑hash revisions so a summary landing mid‑typing can never clobber the user's work.
🚀

Windows · CI/CD · Fleet Management

AI Hub

Packages and delivers the AI applications with one‑click installation, automated updates, and secure delivery to company laptops.

Highlights
  • Built the single entry point for all three tools: a zero‑dependency native window that starts each tool on demand, reuses already‑running instances, and shuts down what it started — chosen over richer UI frameworks because any pip install can be blocked on a managed laptop.
  • Packaged everything into a one‑file Windows installer built in CI, so end users double‑click once with no terminal, Python, or admin steps.
  • Implemented remote fleet control: a single public manifest file drives every installed laptop to update or fully self‑uninstall, with SHA‑256 verification and exact‑version matching so rollback is the same mechanism run backwards — for machines you can never log into.

Independent Projects

Happy Planting Device animation

C++ · Arduino C

Happy Planting Device

Developed the Happy Planting Device using C++ in an Arduino environment, integrating a moisture sensor and customizable MP3 player to enhance household plant care and user experience.

View on GitHub ›
Successorator planner app animation

Java · Android Studio

Successorator (Planner App)

Collaborated on an Android task management app using Java and Android Studio, implementing hierarchical task prioritization and personalized productivity features.

View on GitHub ›
Teaching Computer Science to K-12 Students

Computing Education

Teaching Computer Science to K‑12 Students

Explored critical issues in computing education, with a focus on access and equity. Facilitated conceptual and skill development utilizing fundamental programming concepts such as flow of control, abstraction, and algorithms within a novice programming framework. Implemented innovative computer science teaching methodologies and assessment techniques to foster deep understanding among K‑12 Students.

View Project Site ›