03 · Multi-agent AI
LinkedIn Content Agent.
A three-agent pipeline (Planner, Writer, Reviewer) that turns a topic into a publish-ready post, with OAuth 2.0 token lifecycle management and live reasoning traces.
01 · The problem
A single prompt tends to produce generic posts. Good content needs planning, drafting, and critical editing, and publishing to LinkedIn needs a secure OAuth integration.
02 · What I built
- A three-agent pipeline: the Planner outlines the post, the Writer drafts it, and the Reviewer critiques it and asks for changes.
- Chain-of-Thought prompting so each agent reasons through its step before producing output.
- The LinkedIn OAuth 2.0 authorization code flow with token lifecycle management, so posts can be published from the app.
- A React interface that shows each agent's reasoning trace live, with an approval gate before anything is published.
03 · Key decisions
- 01
Several focused agents instead of one large prompt
Each agent has one job and a short, specific prompt, which makes the output easier to control and each step easier to debug.
- 02
A human approval gate
Nothing is posted without explicit approval, so the automation speeds up writing without taking control away from the author.
- 03
Visible reasoning traces
Showing what each agent decided and why makes the pipeline transparent and helps tune the prompts.
04 · Results
- Topic to publish-ready post in one flow
- A live reasoning trace for every agent
- OAuth-based publishing with an approval step