AZIT
STACK MAP

STACK.

WHAT I WORK IN · READ FAST · HAVEN’T SHIPPED

LANGUAGES
  • Python 3.11
    Primary; the BLUE LAB actor-LoRA pipeline (company work) and personal R&D
  • TypeScript / JS
    Read-only; comfortable ramping for project work
  • Bash
    Daily operational use; Git Bash on Windows
  • SQL (SQLite)
    Job-search DB design, daily queries
ML — TRAINING
  • AI-Toolkit
    LoRA fine-tuning runner; primary training entry point
  • FLUX.2 Klein 9B
    Base model for the BLUE LAB actor-identity LoRA (company work)
  • Z-Image Turbo
    Base model for the synthetic character 'Lee Seol' LoRA (personal R&D, AZIT)
  • PyTorch (via tools)
    Backend; not authored, used through AI-Toolkit / ComfyUI
  • Diffusers concepts
    Familiar with sampling, schedulers, CFG
ML — EVALUATION & ID
  • ArcFace
    Per-step face similarity; primary identity metric
  • InsightFace
    Detection + embedding extraction in evals
  • Parameter sweeps
    Three-round narrowing CFG / Steps / max_shift
  • Overfit monitoring
    Real-time curve watching for early-stop signals
INFERENCE
  • ComfyUI
    Node-graph workflows + HTTP API integration
  • Higgsfield.ai
    Cloud generation; commercial baseline (Soul 2.0) in the BLUE LAB benchmark, dataset images for Lee Seol
  • StabilityMatrix
    Local model and environment manager
LLM / GEN
  • Grok
    Face generation — Jojo Shorts pipeline (production) and the Lee Seol anchor face (text-only)
  • Anthropic Claude
    Daily collaborator; this very portfolio built with assistance
  • OpenAI / Gemini APIs
    Familiar; not currently in production code
  • ElevenLabs V3
    Voice synthesis; production use in Jojo Shorts
  • Synthesizer V Studio Pro
    AI vocal synthesis; production use in audio pipelines
  • RVC voice conversion
    Retrained own voice onto a target convincingly enough for portfolio
INFRA / COMPUTE
  • RTX 4070 (12 GB)
    Local machine; Lee Seol LoRA training (2,502/3,000 steps in 5 h 02 m)
  • RTX 5090 (rented) · RunPod
    Rented / cloud GPU; used for the BLUE LAB actor-LoRA runs
  • Windows 11
    Primary OS; conda env 'facecheck' for InsightFace/ArcFace
  • Cloudflare Pages
    Hosts this portfolio; deployed via Wrangler CLI
  • GitHub
    Source control; public repos: comfyui-workflows, AC-Stock- (Pagard AC work)
DOCS & PROCESS
  • Markdown
    All project documents authored in Markdown first
  • Structured experiment logs
    Per-run JSON / TXT logs feeding analysis scripts
  • Astro + Tailwind
    This portfolio site; static-first, no JS framework lock-in
NOT YET SHIPPED

Honest list of stack items I haven’t shipped to production yet. Read familiarity in most cases; willing to ramp.

  • React
    Read-only; have not shipped production React
  • Node.js
    Read-only; can build small CLIs but not backend services at scale
  • GraphQL
    Read familiarity only
  • Postgres
    SQLite experience translates; not yet shipped Postgres in prod
  • AWS / GCP
    Tutorial-level; have not run production workloads on managed cloud
  • Hasura
    Have not used
  • DevOps / CI-CD
    Basic GitHub Actions familiarity; not yet shipped production pipelines
  • Audit / assurance domain
    New; willing to onboard with structured guidance
← SCRIPTS ABOUT ↗