FORM PR-02 · PROJECT PR-0003 · AI Workflow Engineering

AI News Intelligence & Content Automation

Built an automated news-intelligence workflow that monitors Paraguay immigration and expat developments, uses Claude for relevance analysis and draft content, and organizes traceable results in Google Sheets.

PR-0003.1

The challenge

Scattered sources, relevance, and repeat runs

Discovery: News spread across regional and industry sources
Relevance: Useful residency updates versus general news
Repeat runs: Recurring feeds can surface the same story

Why it matters

Manual review means finding articles, assessing their significance, summarizing them, and recording source details for editorial use. The workflow needed contextual relevance scoring, source traceability, inspectable filtering decisions, and reliable duplicate prevention across scheduled collection runs.

PR-0003.2

How the workflow works

Collection, AI processing, and editorial handoff

Stage	Workflow behavior
01	A scheduled n8n trigger collects articles through RSS and HTTP integrations.
02	Code and merge nodes normalize headlines, source URLs, and publication dates.
03	Claude scores relevance to Paraguay immigration, residency, SUACE, and expat life.
04	Relevant items receive a summary, category, expat-impact note, and draft social post.
05	Google Sheets stores accepted and rejected records; Slack and optional Docs support handoff.

Output · Editorial review
Guidance · Human verified

PR-0003.3

Duplicate prevention

Consistent article identity across runs

Normalize source URLs before comparing incoming stories with existing spreadsheet records.
Normalize headline formatting so presentation differences do not create duplicate entries.
Create deduplication keys from stable article fields rather than relying on raw feed values.
Read existing Google Sheets records before allowing new items into downstream processing.
Filter recurring stories while preserving headline, URL, publication date, and source metadata.
Keep deterministic identity checks separate from Claude's interpretive relevance assessment.
Retain accepted and rejected outputs so filtering decisions remain visible and adjustable.

PR-0003.4

Project coverage

Technology · Outputs · Engineering decisions

Technology

n8n
Anthropic Claude
Google Sheets
RSS Integrations
HTTP Integrations
Code Nodes
Merge Nodes
Google Docs
Slack
Scheduled Triggers
URL Normalization
Dedup Keys
Spreadsheet Routing

Structured outputs

Relevance Score
Category
English Summary
Expat Impact
Draft Social Post
Accepted Tab
Rejected Tab

Engineering decisions

AI for Interpretation
Explicit Data Logic
Source Traceability
Inspectable Filtering
Duplicate Controls
Editorial Review