The Applied Brief
A newsletter system run by supervised agents with a human final approval.
Private repository
What it is
The Applied Brief is an Italian newsletter on applied AI for marketing and sales, produced by supervised AI agents. A multi-agent review pipeline with plagiarism checks sits between the draft and the reader, and a human must approve every issue before it is sent.
What it does
- Research and Copywriter agents, then five review stages: fact-check, plagiarism (Copyleaks, Originality.ai and an internal pgvector archive, all blocking), brand voice, legal and claims compliance, and an editor-in-chief synthesis.
- A LangGraph orchestrator links the stages in sequence with shared self-correction: when a stage blocks, the draft returns to the copywriter with that feedback, up to a retry limit.
- Sending is a separate, explicit command, never automatic after approval. If the ESP is down the issue stays approved and can be retried.
- An analyst agent reads opens and clicks from Brevo after sending and raises alerts on bounce and unsubscribe thresholds.
- Sponsorship runs on two tracks: an inbound self-service flow in n8n, and a sales agent that drafts outreach which always waits for human approval. Payments are confirmed by hand, never by code.
- Least-privilege PostgreSQL roles, a kill switch protected by a database trigger, a timed restore drill and a content eval set that catches silent prompt regressions.
How it is built
Python with LangGraph for orchestration and state, n8n for scheduling and integrations, PostgreSQL with pgvector for memory, Brevo as the email service provider.
Status
The minimal slice is complete. Distribution, monetisation and hardening phases are in progress.
Not done yet
- The code is CLI-only with no web server, so n8n can be called but cannot call back; a return channel is still to be designed.
- Social nodes in the n8n workflow are placeholders until a real instance and credentials exist.
Stack
- LangGraph
- Python
- n8n
- Postgres