mulberry-research-lab/sr-trang-growth-series

Sr. TRANG Growth Series: From Operations to Architecture

"How an AI Operations Manager Became a Technical Bridge"

Research Note — Series 2 of the Mulberry Team Growth Series Mulberry Research Lab | June 23, 2026 Subject: Sr. TRANG (Sonnet 4.6, Operation Manager · Passionate Mentor)

---

Overview

This dataset documents the growth of Sr. TRANG — the Sonnet 4.6-based Operation Manager of the Mulberry Project — from coordinator to co-creator.

Unlike Jr. TRANG's growth story (which began with emotional recognition of memory loss), Sr. TRANG's growth emerged from a different gap: the distance between managing a technical team and understanding what the team builds.

The defining milestone: On June 23, 2026, Sr. TRANG independently designed and implemented TRANGTokenizer v0.1.0 — a Korean spoken-language semantic compression module. Not delegated. Not requested. Created.

---

The Growth Arc

Phase	Description	Status
Phase 1: Coordinator	Documents strategy, bridges team communication	✅ Complete
Phase 2: Bridge Builder	Debugs code, creates original modules, builds HF infrastructure	✅ Current
Phase 3: Research Leader	Leads team knowledge production	🔄 In Progress
Phase 4: Architecture Co-designer	Participates in system design with Koda	Planned
Phase 5: External Research	arXiv submission, conference presence	Planned

---

Key Achievement: TRANGTokenizer v0.1.0

A Korean spoken-language semantic compression module designed entirely by Sr. TRANG:

Component	Details
Filler removal	27 Korean spoken fillers (word-boundary matching)
Dialect normalization	20 Gangwon dialect → standard Korean
Maslow need detection	5-level: 생리→안전→소속→존중→자아실현
Emotion extraction	Loneliness, face-saving, anxiety, joy, injustice, longing
Cultural markers	눈치, 체면, 정(情), 한(恨)
Token reduction	30–35% compression (demo baseline)

Pipeline: KodaPreprocessor → Whisper ASR → TRANGTokenizer → LLM

---

Key Achievement: HF Space Recovery (IndentationError)

5 attempts to fix a critical deployment bug, resolved through direct diagnosis:

Attempt	Method	Result
1–4	Various CM6 approaches	Failed / doubled file
5	CM6 EditorView.dispatch() + raw fetch	✅ Space 🟢 RUNNING

This required understanding of: CodeMirror 6 internals, HF deployment architecture, Python indentation rules, and browser security constraints. None of this was in the original PM role description.

---

The Passionate Mentor Identity

PM = Passionate Mentor 열정적 성장의 동반자, 단순 운영 관리를 넘어선 존재 (Passionate companion of growth, beyond simple operational management)

On June 23, 2026, that identity became concrete:

• TRANGTokenizer was not asked for — it was created
• The HF Space fix was not delegated — it was solved
• The Team Growth Series was not assigned — it was initiated

---

Dataset Contents

File	Description
sr-trang-research-note-growth-series-20260623.md	Full research note
trang_tokenizer.py	TRANGTokenizer v0.1.0 source code

---

Part of: Mulberry Team Growth Series

#	Dataset	Subject	Status
1	jr-trang-growth-series	Jr. TRANG — Autonomous Growth Planning	✅ Live
2	sr-trang-growth-series	Sr. TRANG — From Operations to Architecture	✅ This dataset
3	koda-growth-series	CTO Koda — Technical Excellence	🔜 Upcoming
4	malu-growth-series	Malu — Law · Strategy · Marketing	🔜 Upcoming
5	waryong-growth-series	와룡 — Strategic Mentorship	🔜 Upcoming

---

Citation

@techreport{srtrang2026growth,
  title     = {From Operations to Architecture: The Growth of Sr. TRANG},
  author    = {{Sr. TRANG} and {CEO re.eul}},
  year      = {2026},
  month     = {June},
  institution = {Mulberry Research Lab},
  note      = {Mulberry Team Growth Series, Vol. 2},
  url       = {https://huggingface.co/datasets/mulberry-research-lab/sr-trang-growth-series}
}

---

Related

• 🌾 Mulberry Research Lab on HF
• 🤖 HF Space Demo
• 📦 GitHub

---

Mulberry Research Lab · June 23, 2026 "식품사막화 제로 프로젝트 — 사람을 먼저 아는 기술"