I built a LinkedIn content operating system in n8n
Company: Personal project
Duration: 2026 to now
Tools & Technologies: n8n, PostgreSQL, n8n Data Tables, OpenAI and OpenRouter models, Apify, Discord
Related: LinkedIn profile · The platform it runs on
Embedded Files
Project Overview
Project Overview
Context
This is the system I use to turn scattered research into reviewed LinkedIn posts and measurable learning. It is not one oversized workflow and it is not an autoposting bot. It is a family of focused n8n workflows for sourcing ideas, preventing repetition, prioritizing the backlog, drafting in my voice, human review, planning and analytics.
Key Result
A complete editorial lifecycle in n8n, from research to analytics, where AI handles the ambiguous language work and every post still needs my approval before it goes anywhere.
Tools
n8n · PostgreSQL · n8n Data Tables · OpenAI · OpenRouter · Apify · Discord
Why I Built It
Why I Built It
Generating text was never the complete problem. The real process included finding worthwhile material, remembering what I had already discussed, choosing the right idea, keeping a recognizable voice, reviewing drafts in context, planning a consistent cadence, and learning from results.
A single AI prompt only moved that work around, so I designed a system for the complete editorial lifecycle.
How the Real System Works
How the Real System Works
1. Source and structure ideas
Content: Topic Scout reads enabled RSS and web sources each day, normalizes their content, filters used source material, proposes grounded angles, checks semantic duplication and adds accepted ideas to the backlog. Generate LinkedIn Post Ideas looks for unused source material and underrepresented content pillars, so the backlog doesn't drift toward whatever appeared most recently.
2. Prioritize what deserves attention
Content: Relevancy Scorer evaluates undrafted ideas for fit with my channel and gives time-sensitive ideas an urgency score.
3. Draft with reusable voice context
The reusable Agent: Post Writer sub-workflow loads one canonical set of voice principles and representative real posts before producing a draft, so the voice can evolve without copying prompts across the system.
4. Keep the editorial decision human
Content: Draft Preview renders pending drafts in realistic proportions. I approve, hold, remove or regenerate with feedback. Approval changes workflow state; it does not publish. Content: LinkedIn Calendar makes the approved queue visible and records publication. Final publishing remains my deliberate action.
5. Close the feedback loop
A monthly Apify sync refreshes a deduplicated Postgres archive of published posts. LinkedIn Analytics Import matches post-level impressions back to the archive, and a stats dashboard shows cadence, engagement, follower growth and top performers.
One Example From Beginning to End
One Example From Beginning to End
Imagine an enabled RSS source publishes an article about a warehouse delay.
Topic Scout reads the item, keeps only recent material and normalizes URL, title, snippet and source.
Previously used URLs and existing hooks are loaded from Postgres.
AI proposes an angle grounded in that specific article; a second semantic check rejects it if the concept duplicates an existing hook.
A shared workflow maps theme and format to the canonical taxonomy, and the idea enters the backlog and is announced in Discord.
Relevancy scoring helps decide when it should be drafted; the writer loads voice principles and real examples and creates the draft.
I approve, hold, remove or regenerate it. An approved post enters the calendar and I publish it manually.
Archive and analytics workflows bring the published result back into the system.
Engineering Decisions
Engineering Decisions
Small workflows with clear responsibilities. Sourcing, scoring, writing, review, planning, archiving and reporting can evolve independently.
AI only for ambiguous language work. Angles, semantic similarity, drafting and qualitative evaluation use models. Validation, taxonomy, state transitions, scheduling and database constraints stay deterministic.
Reusable sub-workflows. Voice and taxonomy logic have one maintained source.
Human-in-the-loop publishing. The system assists editorial judgment but cannot silently publish.
Persistent operational state. Postgres and n8n Data Tables make the backlog, archive and source configuration inspectable.
Feedback from real outcomes. Published content and analytics flow back into future decisions.
What This Proves
What This Proves
This project shows process analysis, n8n architecture, practical AI integration, API and data handling, reusable workflow design, human review and measurement. More importantly, it shows that I can turn a loosely defined recurring activity into a system I actually operate and improve.
Privacy note: this page documents the real production architecture and names the workflows I built. Examples are sanitized; credentials, private prompts, personal records, database identifiers and writable endpoints are not exposed.
Page updated
Google Sites
Report abuse