LinkedIn Agent

I wanted to create a blog, and to automatically export the content to LinkedIn for a one input/multiple outputs system.

Blog

Using a sveltekit blog template and postgres, I deployed the website to my kubernetes cluster. The great advantage of this template was the incredibly easy content modification directly from the website, eliminating the need for CMS or repo update upon new blog posts creation.

I then simply added a webhook on blog creation, directed toward a n8n workflow.

N8N

I created a workflow which takes in a slug directly in the url, then fetches the blog’s content from the database. This simplified the connection between the blog and n8n, requiring only the URL, the slug, and basic security.

An AI agent (I used deepseek r1 for its great reasoning capability) picks up the content, and creates a teaser for linkedin. Using human in the loop tools, the AI agent first validates with me the content, then sends it directly to LinkedIn.

Conclusion

As simple as it gets, getting the system prompt right was a challenge. AI is sometimes like a dumb child, sometimes like an utter polite colleague. Getting it to respond correctly takes time. The other parts of this system are quite simple, and I really enjoy the usefulness alongside the independence between the parts.