AZIT
FACE 02 / ENGINEER ARTIFACT-FIRST PIPELINES

JUNGCHAN LEE

ML SYSTEMS ENGINEER

Self-taught ML engineer. I design and run Python pipelines for LoRA fine-tuning, parameter optimization, and identity evaluation (code written with Claude Code; design, runs, and verification are mine). At The Blue Studio’s BLUE LAB (staff, AI content production, March 2026) I trained identity LoRAs on real actor photography and benchmarked them with ArcFace against a commercial baseline (Higgsfield Soul 2.0) — the documents, nine scripts, and numbers are published with the company’s permission. My personal R&D (AZIT) is a Z-Image Turbo LoRA of ‘Lee Seol’, a synthetic character whose face was generated from text only. Now automating short-form and chart video production with Claude Code at Pagard AC, checked against evaluation data.

KEY METRIC · KEY RESULT
0.6050 ArcFace  ·  vs 0.5090 baseline
The Blue Studio BLUE LAB (company work) · actor LoRA on FLUX.2 Klein 9B · published with permission  · +18.9% OVER BASELINE
MOST RECENT ROLE

Video Automation

Pagard AC — short-form & chart video

PERIOD
August 2026 — present
SCOPE
Corpus · judge eval · rulebook QA · edit automation
STACK
Python · Claude Code · After Effects · Premiere
STATUS
Open to roles
SEE THE WORK  ↗
WORKING ENVIRONMENTS

Tools that ship every dependency

How my personal R&D (AZIT) ‘Lee Seol’ LoRA runs — the dataset preparation step in AI-Toolkit, and the Z-Image Turbo inference graph in ComfyUI. Full step-by-step trace in VISUAL WORKFLOW ↗.

ComfyUI inference workflow — LoRA load, dual CLIP encode, KSampler, FaceDetailer (Impact Pack)
ComfyUI inference workflow (personal R&D · Lee Seol LoRA, Z-Image Turbo) — LoRA load, dual CLIP encode, KSampler, FaceDetailer
HONEST FRAME
B.A. in Digital Content (DCDC, 2025), self-taught ML engineering through company LoRA work at The Blue Studio BLUE LAB and my own R&D. I do not have years of ML-board-backed experience; I do have evidence I can scope, build, and measure a non-trivial ML pipeline — nine scripts (written with Claude Code, designed and verified by me), a benchmark, a structured sweep, a visual workflow trace — doing the talking instead of credentials.
OTHER FACE
AUDIO FACE  ↗

K-Pop Producer & Audio Engineer

Solo K-Pop songwriter, beat-maker, mixing & mastering engineer. End-to-end in Cubase Pro and FL Studio. Master delivery −10.5 LUFS, +6.0 LU range.

↗