Hyunae Park

AI Engineer Seoul, South Korea

hyunaeee@gmail.com

I build AI services around real workflows, from retrieval and agent tools to deployment, evaluation, and ongoing maintenance. My work spans language, speech, computer vision, and interactive products.

Experience

LIKE Corporation

Meeting Assistant

  • Own product planning, development, deployment, operations, and maintenance of an internal meeting assistant.
  • Built a FastAPI and React workflow for transcription, speaker diarization, structured minutes, and Notion/email delivery using faster-whisper, pyannote, Claude, and an RTX 4090.
  • Separated long-running jobs from HTTP requests, added reconnection after lost upload responses, and preserved transcripts and minutes when downstream steps failed.

MED-RAG · client installation

  • Installed a local RAG assistant on the personal PC of a professor treating breast cancer at Korea University Anam Hospital, and continue to provide remote updates.
  • Connected case and guideline retrieval with visible source evidence. The installation is scoped to the professor’s PC; public evaluations use a separate synthetic dataset.

NC AI

AI Data Division.

SolverX

Startup development work; the company operated as VoltWin during my employment.

LightMomentus

Contributed to software development as an early member of the startup.

Selected Projects

Terracotta

  • Implemented model routing, OpenAI and Anthropic tool loops, HTTP MCP connections, and execution traces.
  • Added approval and resume flows for calls classified as write actions in the agent loop, plus Docker self-hosting and GitHub Actions container publishing.
Source

MED-RAG · agent evaluation

  • Built a researcher–reviewer agent sequence with case and guideline search tools; evaluated retrieval and answer quality separately.
  • Published a 24-question synthetic evaluation: gold-document retrieval on 23/24; an evidence-aware LLM judge passed 20/24 answers.
  • Investigated fabricated citations and over-refusal in QLoRA experiments, and recommended the base model after tuning reduced helpfulness.
Evaluation

MED-RAG Serving Lab

  • Pinned model, prompt, dataset, and runtime versions; implemented an SSE gateway, quality checks, and release-artifact validation.
  • Published 14 rehearsal checks covering HTTP behavior and release gates with synthetic controls and a scripted CPU upstream. GPU inference and production rollout/rollback remain outside the verified scope.

Smart Docent

  • Lead product planning, priorities, scheduling, and role allocation for a location-aware AI guide.
  • Own the place-data pipeline and contribute to architecture and shared LangGraph agent implementation, with Mapbox and multilingual guidance.

VisionEye

  • Compared detection and tracking configurations on generated-video inputs, separating observed behavior from unmeasured real-world accuracy.
  • Built a CPU model API and verified worker recovery after injected timeouts and process exits.

RoboSkill Lab

  • Compared one-shot pushing with repeated observation and re-approach across 24 simulated runs, recording success, task time, contacts, and failures.

Research

Korea Brain Research Institute

  • Developed experimental software and GUIs and contributed to CNN analysis and DeepLabCut-related work using MATLAB, C#, and Python.

Korea Brain Research Institute

  • Developed a JavaScript application for delivering behavioral experiments.

Teaching

Beyond Coding

  • Taught computer science and programming in English to international-school students.

T&B School Edu

  • Delivered computer science career-exploration classes for high-school students across Korea.