2nd/10_Wiki/Topics/AI_and_ML/Connect-AI-Documentation.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-connect-ai-documentation	Connect AI Documentation	10_Wiki/Topics	verified	self	ConnectAI documentation AI agent documentation system docs SOP API spec	none	B	0.83	applied	documentation ai-agent sop api-spec knowledge-base project-internal		2026-05-10	pending	language applicable_to documentation Internal AI Agent Project SOP Onboarding

Connect AI Documentation

매 한 줄

"매 complex AI 의 readable 의 manual". 매 system architecture + SOP + API spec 의 unified. 매 modern: 매 living doc + 매 RAG 의 query-able + 매 AI 의 self-update. 매 Codebase_Onboarding_Guide 와 의 align.

매 핵심 component

1. System Architecture: 매 C4 model.
2. SOP: 매 step-by-step operation.
3. API Spec: 매 OpenAPI / GraphQL schema.
4. ADR: 매 decision record.
5. Glossary: 매 term.
6. Runbook: 매 incident response.
7. Onboarding: 매 new dev.

매 living doc 의 element

• CI 의 sync: 매 stale 의 detect.
• RAG-queryable: 매 LLM 의 answer.
• Code → doc: 매 generate from source.
• Multi-format: markdown + diagram + interactive.

매 best practice

• 매 README first.
• 매 Diátaxis (tutorial / how-to / reference / explanation).
• 매 minimum viable doc.
• 매 example > prose.
• 매 versioning.
• 매 link > duplicate.

💻 패턴

Doc-as-code (Mintlify / Docusaurus / Astro)

docs/
├── index.mdx
├── architecture.mdx (C4)
├── runbook/
│   ├── incident-database.mdx
│   └── incident-deploy.mdx
├── api/
│   └── (OpenAPI generated)
└── onboarding.mdx

Living doc (CI sync check)

- name: Doc drift check
  run: |
    npm run extract-types
    git diff --exit-code docs/api-types.json || \
      (echo "Doc out of sync — regenerate" && exit 1)

RAG over docs

from langchain.vectorstores import Chroma
from langchain.text_splitter import RecursiveCharacterTextSplitter

splitter = RecursiveCharacterTextSplitter(chunk_size=1000)
chunks = splitter.split_documents(load_md_files('docs/'))
vectordb = Chroma.from_documents(chunks, embeddings, persist_directory='./doc-rag')

def ask_doc(question):
    relevant = vectordb.similarity_search(question, k=5)
    return llm.generate(f"Use these docs:\n{relevant}\n\nQuestion: {question}")

OpenAPI generation

// 매 zod-to-openapi
import { extendZodWithOpenApi } from '@asteasolutions/zod-to-openapi';
extendZodWithOpenApi(z);

const userSchema = z.object({
  id: z.string().openapi({ example: 'uuid-...' }),
  email: z.string().email().openapi({ example: 'user@example.com' }),
}).openapi('User');

🔗 Graph

• 응용: Codebase_Onboarding_Guide · Codebase_Maps_and_Interactive_Tours · ADR · C4_Model
• Adjacent: Asset-Specific-Knowledge · Software Architecture Styles

🤖 LLM 활용

언제: 매 internal AI project 의 doc design. 언제 X: 매 single-file script.

❌ 안티패턴

• Stale doc: 매 CI 의 sync 의 X.
• Single huge file: 매 navigate 의 X.
• No examples: 매 pure prose.
• Code 의 doc 의 mismatch: 매 source-of-truth 의 ambiguous.

🧪 검증 / 중복

• Verified (Diátaxis, doc-as-code best practice).
• 신뢰도 B.
• Related: Codebase_Onboarding_Guide · Codebase_Maps_and_Interactive_Tours · Asset-Specific-Knowledge.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — components + 매 doc-as-code / RAG / OpenAPI code