My role: System Architect & Builder

Built an end-to-end content automation system in n8n that monitors fresh business news from RSS feeds and Reddit, filters relevant stories using keyword logic and low-cost LLM checks, parses full articles, extracts key facts, scores relevance, generates draft posts, and transfers only high-scoring results into a final review table.

The system integrates OpenAI, Claude, Google Sheets, Make, Buffer, and Telegram alerts to create a stable, low-touch workflow that saves significant marketer time every week and keeps the content pipeline consistently filled.

Skills and deliverables

• n8n
• Google APIs
• Make.com
• Web scraping
• OpenAI API

Overview

I designed and built an end-to-end content automation system that turns fresh AI industry news into ready-to-review social media drafts for LinkedIn, X, and Threads.

The goal was simple: make sure the marketing team always had a steady flow of relevant, timely content every week without spending hours searching for topics, reading articles, and drafting posts manually.

The goal

The main challenge was to create a reliable system that could continuously supply fresh content ideas and draft posts for multiple social media channels.

Before this workflow, marketers had to:

• manually search for news
• review multiple sources
• decide what was relevant
• summarize articles
• write posts from scratch

This process took too much time and made content production inconsistent.

What I built

I created a multi-step automation system in n8n that monitors fresh AI-related news, filters it, extracts key facts, scores relevance, generates social media drafts, and sends only the best results into a final posting table.

The system combines:

• Google Alerts and RSS feeds for source monitoring
• Reddit parsing for additional content signals
• OpenAI models for filtering, summarizing, scoring, and drafting
• Claude for making posts sound more natural
• Google Sheets as a control and status layer
• Make and Buffer for final publishing
• Telegram alerts for error monitoring

How the system works

1. Source monitoring

I started by creating targeted Google Alerts queries focused on AI applications and related topics.

These alerts generated RSS feeds, which were then monitored by n8n every morning.

At the same time, a separate workflow parsed selected Reddit threads to capture additional relevant discussions and ideas.

2. First relevance check

When new items appeared, the system checked them in two stages:

• first through keyword matching
• then through a low-cost OpenAI model to confirm whether the news was actually relevant

Relevant items were saved into a source table.

3. Article parsing and fact extraction

Another workflow then visited each source article, parsed the actual page content, and extracted the core facts from the full text.

This step turned long-form articles into short, structured summaries that could be reused later in content generation.

4. Scoring

The summarized news items were then scored on a 1 to 10 relevance scale based on how well they fit the brand’s content goals.

These scores were written back into the table, making it easy to track what was worth using.

5. Post generation

Only items with higher scores moved forward.

From there, the system generated:

• a LinkedIn post
• an X post
• a Threads post

This stage was not a single prompt.

I split it into multiple steps:

• one workflow generated the initial draft
• another step used Claude to make the copy sound more human
• a memory-based layer injected brand or marketer-specific notes and context

All generated versions were written into structured tables for review.

6. Final quality filter

A final scoring workflow reviewed the generated posts and evaluated their quality.

Only posts that scored 7 to 10 were transferred into the final posting table.

7. Publishing pipeline

From the final table, approved content could then be sent through Make and Buffer for publication across social channels.

This meant the system did not publish raw output directly.

It prepared reviewed, structured drafts and moved only strong candidates into the final posting stage.

Reliability and control

I designed the workflows as separate time-based automations, so one failure would not break the whole system.

If one workflow failed because of rate limits, parsing errors, or server-side issues, the rest of the pipeline could continue running independently.

I also set up:

• status tracking in tables
• error alerts to Telegram
• clear step-by-step visibility into whether a news item had been found, parsed, scored, drafted, reviewed, or transferred

This made the system easy to monitor and maintain.

Results

This automation created a reliable weekly content pipeline and significantly reduced manual work for the marketing team.

Business outcome

• marketers no longer had to search for news manually
• the team always had relevant content ideas and draft posts ready for review
• content production became more consistent and easier to manage

Time savings

I estimate the system saved the marketer at least 7 hours per week, mainly by eliminating manual research, summarizing, and first-draft writing.

Workflow outcome

Instead of creating content from scratch every time, the marketer now reviews prepared drafts once a week, makes edits where needed, and moves forward with publishing.

That drastically reduced repetitive work while still keeping human control over the final quality.

My role

End-to-end System Architect & Builder.

I designed the workflow logic, built the automation architecture, connected the tools, created the filtering and scoring stages, structured the data flow through tables, added monitoring and alerts, and made the overall system resilient enough for daily use.

Tech stack

• n8n
• OpenAI API
• Claude
• Google Alerts
• RSS feeds
• Reddit parsing
• Google Sheets
• Make
• Buffer
• Telegram alerts