n8n Workflow Automation · AI-Driven Content Curation & Publishing
Designed and deployed a production-grade, multi-stage automation pipeline using n8n that operates on a 2-hour schedule. The system ingests high-volume technical news, normalizes the data, detects duplicate coverage, scores content via Claude 3.5, and generates publication-ready drafts pushed directly to Beehiiv upon Slack approval.
The Challenges
- Manual curation consumed hours of editorial time daily
- Dozens of scattered sources made consistent coverage impossible
- No scalable way to score, rank, and draft on-brand
Solutions
- Designed an hourly automated pipeline ingesting sources
- Built semantic de-duplication using embeddings and a 30-day rolling vector memory
- Implemented LLM-based relevance scoring against a weighted editorial rubric
- Automated per-section drafting (Big Story, Roundup, Tools, What-to-Watch) in brand voice
- Added a fact-consistency verification pass with verbatim rules for CVEs, figures, and dates
- Integrated a human-in-the-loop approval gate via Slack before publishing
- Connected directly to Beehiiv API to stage schedulable newsletter drafts
Key Outcomes
- Reduced editorial turnaround from hours to minutes per issue
- Eliminated duplicate/low-relevance content through automated scoring
- Preserved editorial control via mandatory human approval step
- Delivered a fully configurable, scalable pipeline — adding sources requires no re-engineering
Technical Architecture
Workflow Automation:
n8n
AI & Machine Learning
Anthropic Claude API, VoyageAI Embeddings
Database
PostgreSQL (pgvector)
Integrations & Publishing
Beehiiv API, Slack Webhooks & Block Kit
Data Sources:
Google Sheets
Core Architecture & Key Features
Multi-Source Hybrid Ingestion
Fetches real-time feeds every 2 hours from diverse API/RSS sources and dynamically enriches, parses, and normalizes incoming metadata and full-text content.
Vector Embedding & Deduplication
Generates text embeddings for all incoming articles. Stores embeddings in a PostgreSQL database with pgvector to identify and remove duplicate coverage across different sources.
Multi-Dimension LLM Content Scoring
Evaluates deduplicated items using Claude across six specific dimensions and Commercial Impact, Audience Suitability, and Novelty. Automatically calculates a weighted score and shortlists top stories.
Automated Newsletter Drafting
Dynamically formats content according to strict brand voice guidelines, enforcing verified figures, accuracy, and structured section headings. Assembles semantic HTML bodies complete with source references and callouts.
Human-in-the-Loop Approval & Publishing
Posts interactive Slack blocks with full content summaries, fact-check alerts, and action buttons (Approve, Reject). Upon one-click approval via Slack webhooks, automatically creates draft posts inside the Beehiiv publication API.
Observability & Error Handling
Integrated Slack error triggers for instant pipeline failure alerts. Automated daily heartbeat reports summarizing collection metrics, top similarity scores, and duplication counts.
Let's Design Your Website
From concept to launch, I create websites that are visually stunning, user-friendly, and conversion-focused. Combining modern design principles with seamless functionality, I make sure every interaction delights your visitors and drives results.