Multi-agent content pipeline
Sole builder — prototype
Three agents (research, strategy, content) turn signals from GitHub, ArXiv and O'Reilly into platform-ready posts. Local Ollama; nothing is scheduled until a human approves it.
Architecture
Problem
Research signals need to become platform-ready posts without API spend, and without publishing until a human approves.
What I built
Signals from GitHub, ArXiv and O'Reilly feed a research agent, then a strategy agent, then a content agent. Output waits in a FastAPI dashboard. A human must approve before anything is scheduled. Local Ollama. Versioned prompt templates.
Engineering decisions
- 01
Three specialised agents (research, strategy, content) rather than one mixed prompt.
- 02
Local Ollama keeps API cost at zero.
- 03
Nothing is scheduled until a human approves it in the FastAPI dashboard.
- 04
Prompt templates are versioned. Evals are in progress.
Outcome
Prototype. Local Ollama at zero API cost. Nothing is scheduled until a human approves it. Versioned prompt templates; evals are in progress. Private repo.
