class AIEngineer:
name = "TAE YOU KIM (shoman2)"
focus = ["AI Agent Systems", "LLM Harness Engineering", "Workflow Automation"]
stack = ["Python", "Claude SDK", "JavaScript", "pandas", "sklearn"]
current = "Building production-grade multi-agent exam authoring harness"
philosophy = "Ship automated pipelines that replace 100x manual effort"AI Agent Harness μ λ¬Έ μμ§λμ΄. 볡μ‘ν μ
무 νλ¦μ μ€μΌμ€νΈλ μ΄ν° β μ λ¬Έ μμ΄μ νΈ β κ²μ¦ 루ν κ΅¬μ‘°λ‘ μλνν©λλ€.
νμ μμ λΆΒ·ν΄μμμ°μΈμ¬κ°λ°μ λ± κ³΅κ³΅κΈ°κ΄ λμ LLM κΈ°λ° μλν μμ€ν
μ€κ³/ꡬμΆ/컨μ€ν
κ²½ν 보μ .
κ³΅κ³΅κΈ°κ΄ AI μ±νΌμΈ μν 30μΈνΈ μλ μΆμ νλ€μ€ β κΈ°κ΄ μ²¨λΆ μλ£λ§ λ£μΌλ©΄ 90κ° λ¬Έμ μ§Β·ν΄μ€Β·μ±μ μλ²Β·PDFΒ·CBT JSONκΉμ§ μμ μλ μμ±
[Input] κΈ°κ΄ μ 곡 μ²¨λΆ μλ£ (보λμλ£ Β· λ²λ Ή Β· ν΅κ³ Β· CSV)
β
βΌ
[Round 0] attachment-analyzer β μλ£ λΆλ₯ & μ½ν
μΈ μ ν μλ³
[Round 1] set-designer β 30μΈνΈ Γ 3κ³Όλͺ© μ£Όμ 맀ν
[Round 2] question-writer Γ90 β λ¬Έν 5κ° + μλλ¦¬μ€ (λ³λ ¬)
[Round 3] solve-writer Γ90 β solve.py + λ΅μμ§ + ν΄μ€
[Round 4] leak-detector β μ λ΅ λ
ΈμΆ 8-pattern LLM μ§μ κ²μ¬ βΊ
[Round 5] solve-verifier β solve.py β λ΅μμ§ μ ν©μ± κ²μ¦ βΊ
[Round 6] builder β μμμ/μ±μ μ PDF Β· zip Β· CBT JSON
β
βΌ
[Output] 30μΈνΈ Γ 3κ³Όλͺ© = 90 λ¬Έμ μ§ ν¨ν€μ§ (μμ λ°°ν¬ κ°λ₯)
| νλͺ© | μ€ν |
|---|---|
| μμ΄μ νΈ κ΅¬μ± | μ€μΌμ€νΈλ μ΄ν° 1 + μ λ¬Έ μμ΄μ νΈ 6 + μ€ν¬ 4 |
| λ¬Έν μλ μμ± | κ·Έλ¦°(AIΒ·μμ ) / λΈλ£¨(PythonΒ·ML) μ΄μ€ νΈλ |
| νμ§ λ³΄μ¦ | μ λ΅ λ ΈμΆ ν¨ν΄ 40+κ° μλ κ°μ§, solve.py μ€ν κ²μ¦ |
| 보μ | μμμΒ·μ±μ μ νΈλ¦¬ λΆλ¦¬, λμΆ 0건 κ²μ¦ νμ |
| λ°°ν¬ μ°μΆλ¬Ό | PDF Β· μ²¨λΆ ZIP Β· CBT JSON Β· μ±μ λꡬ |
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Orchestrator Layer β
β (λΌμ΄λ κ°λ
Β· μ¬μμ± λ£¨ν Β· μ¬μ©μ νμΈ) β
ββββββββββββββ¬βββββββββββββ¬βββββββββββββ¬βββββββββββββ¬ββββββββββ€
β Analyzer β Designer β Writer β Verifier β Builder β
β (λΆλ₯Β·μλ³) β (μ€κ³Β·λ§€ν) β (λ¬Έν μμ±) β (κ²μ¦Β·λ£¨ν) β (λΉλ) β
ββββββββββββββ΄βββββββββββββ΄βββββββββββββ΄βββββββββββββ΄ββββββββββ€
β Skills / Domain Knowledge β
β exam-authoring Β· leak-detection Β· solve-verification β
β Β· difficulty-tuning β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
ν΅μ¬ μ€κ³ μμΉ:
- LLMμ΄ μ§μ read β μ κ·μ grepμΌλ‘ μ‘μ μ μλ μλ―Έμ μ λ΅ λ ΈμΆ νμ§
- μ¬μμ± λ£¨ν β κ²μ¦ μ€ν¨ μ ν΄λΉ λΌμ΄λλ‘ μλ 볡κ·
- μμμΒ·μ±μ μ νΈλ¦¬ μμ λΆλ¦¬ β μ±μ μλ£ λμΆ 0-tolerance