Capstone, a running automation process

Level 3 showed what automation is, how Zapier works, how n8n compares and how webhooks tick. Now you build your own automation that runs every day or every week for you. It's the first real mini-employee in your portfolio.

The task

Pick a recurring process you already do by hand and automate it. Examples:

• Pull new jobs from two job portals every day at 8am into a Google Sheet, with a score for how well they fit.
• Sort incoming emails to a specific address automatically by category and push to Notion.
• Every Friday assemble a summary of your week's activities from multiple sources (calendar + todo app + mail count) into a Markdown report.
• For every new invoice in your "Inbox" folder: run OCR, extract the data, enter into a spreadsheet.

What the system has to do

• Run without you pressing a button. Time trigger or event trigger.
• Talk to at least one external service (API or webhook or both).
• With an AI step inside. Zapier AI Action, OpenAI node in n8n, an LLM call via HTTP node, doesn't matter how, but AI is part of the chain.
• Error case handled. What happens when the API is down? What if the AI model returns an unexpected output? What if too many entries come in?
• At least five runs done. Not just one "it worked" but you observe how it behaves over days/weeks.

What you should produce

1. The automation itself. Zapier or n8n, live and running.
2. A diagram (hand-drawn photographed, miro/draw.io, doesn't matter) showing the flow: Trigger, Step 1, Step 2 ... Output.
3. A readme.md explaining:
   • What the process does.
   • Which services are involved and why these.
   • How often it runs and why this interval.
   • What happens in the error case.
   • What surprised you while building it.
4. Log or screenshot of at least five successful runs.

What's strong

The process really saves you time. The readme.md is written so someone else could rebuild it. You provoked at least one error case on purpose and saw how the workflow reacts. You know what the process costs per month in Zapier/n8n resources.

What's weak

A tutorial rebuilt from a YouTube video. Zero error handling, fails on the first edge case. Output is filed nowhere, evaporates into nothingness. No measurement whether it actually helps.

Bonus

Build in a Human-in-the-Loop step (see Level 5 Lesson 5). For every non-trivial decision the automation puts a draft in your inbox or a queue, instead of executing directly. You accept with a click.

That's the leap from "automation that runs" to "automation you trust".

Where it belongs

Folder portfolio/level-3-capstone/. This is the first portfolio piece where you can show a company that you don't just understand automation conceptually but use it productively. The readme.md is more important than the Zapier workflow itself, because companies hire you for your understanding, not for your tool.

Time

90 to 120 minutes of active work. Plus a week passively while the process runs. At the end of the week another 30 minutes for evaluation and readme.

Wrap-up and paywall

Take the Level 3 checkpoint. After that you're at the edge: the three free levels are done, from Level 4 it goes into Memory + MCP + Multi-Agent. The gap between Level 3 and 4 is the biggest qualitative leap in the academy. What you've learned so far, many can. What comes from Level 4 on, few can.

If you study seriously and don't drop out, in Level 6 you'll publish your own MCP server.