Optimizing AI-Assisted Development: Configuring Context for K-Beauty Export & Localization

Key Takeaways (핵심 요약)

1. Context is King: 단순한 코드 룰을 넘어, 비즈니스 목표(HSK 매핑, $0.01 단위의 가치)와 제품 철학(초월 번역)을 명시한 project_rules.md를 도입하여 AI의 이해도를 비약적으로 높였습니다.
2. Safety First: FDA 규제 준수(Whitening vs Brightening)와 같은 민감한 도메인 지식을 규칙 파일에 통합하여, AI가 안전하고 규제 준수적인 결과물을 내놓도록 가이드라인을 세웠습니다.
3. Communication Protocol: 창업자(Founder)와의 소통 효율을 위해, 코드는 영어로 작성하되 '생각의 흐름(Thinking Process)'은 한국어로 기술하도록 하여 의사결정의 투명성을 확보했습니다.

What is the most efficient way to align AI with Business Logic?

In the era of AI-assisted coding, the quality of output is heavily dependent on the context provided. For Beauty Insight Lab, a standard linting configuration wasn't enough. We deal with complex domain logic like HSK Codes for export tracking and nuanced Transcreation rules for US market localization.

To bridge this gap, we deprecated the generic claude.md and authorized a new system prompt file: project_rules.md.

The Anatomy of project_rules.md

We didn't just dump code snippets. We injected the "Soul" of the project into the configuration.

# specific rules integrated into context
# backend/src/translation_agent/rules.py
 
LOCALIZATION_MAPPING_TABLE = {
    "Texture": {
        "쫀득한": ["Fast-absorbing & Satin Finish", "Avoid 'Sticky'. Emphasize absorption."],
        "물광": ["Radiant Glow", "Avoid 'Dewy' alone if it implies 'Oily'."]
    },
    "Marketing": {
        "미백": ["Brightening / Radiance", "NEVER use 'Whitening' (Banned)."],
        "주름 개선": ["Diminishes fine lines", "Avoid 'Anti-wrinkle' (Drug claim)."]
    }
}

Why did we rename it to project_rules.md?

Initially named prompts/claude.md, we realized that this file serves as more than just a prompt for a specific model. It is the Living Documentation of our project's rules.

By moving it to prompts/project_rules.md, we established a centralized "Knowledge Base" that any agent (Gemini, Claude, or custom LLMs) can reference to understand:

1. What we are building (K-Beauty Platform).
2. How we build it (Tech Stack: Next.js 16, FastAPI).
3. Why we build it (To solve the Texture Gap & Compliance issues).

How does this enhance our workflow?

1. Reducing Context Switching

Developers and AI agents no longer need to hunt for "Which HSK code was for Mask Sheets?" inside hsk.ts. The rule is explicit in the context.

2. Preventing "Hallucinations" in Marketing Copy

By explicitly defining the Global Ban List (e.g., prohibiting "Atopy" medical claims), we ensure that generated marketing copy is safe for deployment without heavy manual review.

3. "Co-founder" Mode

We instructed the AI to act as a Co-founder. This means:

• Language Strategy: Code in English, but Thinking Process & Architecture explanations in Korean.
• Proactive Feedback: If a requested feature increases technical debt without business value, the AI is empowered to push back.

This project is part of our journey to expand K-Beauty globally using AI.