Blog
LangGraph + HITL
주제와 조건을 입력하면 여러 에이전트가 병렬로 조사하고, 쓰고, 검토한다 — 품질이 미달이면 interrupt()로 멈추고 사람의 판단을 기다린다.
Built a multi-agent pipeline that generates a lecture plan from a 6-field input (topic, audience, duration, delivery method, tools, constraints) using LangGraph + FastAPI + Gemini API + Tavily.
The pipeline runs in 3 rounds: sequential planning → parallel web research → sequential writing → …
AgentRead more →
Multi-Agent(HITL) with n8n
주제와 조건을 입력하면 AI가 조사하고, 쓰고, 세 관점으로 검토한다 — 품질이 미달이면 Slack으로 사람에게 판단을 넘긴다.
Built a multi-agent pipeline that automatically generates a lecture plan from a 6-field webhook input (topic, audience, duration, delivery method, tools, constraints). n8n orchestrates the full workflow across 20 nodes; a Flask server handles all Gemini and Tavily calls.
The pipeline runs in 3 rounds: …
AgentRead more →
Single vs Multi-Agent
에이전트를 여러 개로 쪼갠다고 항상 더 똑똑해지는 건 아니다
Before scaling an agentic workflow into a multi-agent pipeline, it's worth asking whether the added complexity actually pays off in token cost.
Single-agent overhead accumulates vertically — context grows turn by turn, but prompt caching keeps reuse efficient. Multi-agent overhead spreads horizontally — each sub-agent …
AgentRead more →
Multi-Agent with Claude Code
주제와 조건을 입력하면 9개 서브에이전트가 조사하고, 쓰고, 검토하고, 고친다 — 코드 없이
Built a multi-agent pipeline that automatically generates a lecture plan from a natural language prompt. No code — orchestrated entirely through CLAUDE.md (entry point only, 3 lines) and agent definition files (.claude/agents/*.md).
The pipeline runs in 4 rounds: sequential planning → parallel web research (up …
AIRead more →
ReAct Research Agent with Lang
n8n 캔버스에서 화살표로 그렸던 루프를, 코드 한 줄의 그래프 선언으로 옮겼다.
Reimplemented the same ReAct (Reasoning + Acting) research agent from the n8n
version, this time as an explicit graph in LangGraph. Where n8n expressed the
loop as a line drawn back to an earlier node on a canvas,
add_conditional_edges expresses the same logic in code. …
Read more →