Multi-Agent Content Pipeline with a Critic
Research, draft and critique as separate agents — and prove the critic earns its cost.
Advanced16–26 hours estimatedMulti-Agent
The problem
Multi-agent systems are usually built because they sound sophisticated, then quietly cost three times as much as one good prompt for no measurable gain. This project builds one properly and — the important part — measures whether the extra agents actually improved the output.
Who uses it: Producing drafts on technical topics where accuracy matters more than speed.
Before you start
- Comfortable building single agents
- Python
- A budget for experiments
Stack
PythonLangGraph or a hand-rolled state machineA search APIA judging LLM