--- title: "Capstone, building a real system out of prompts — StudioMeyer Academy" description: "Not single prompts. A reusable, documented chain of tasks." canonical: "https://studiomeyer.academy/en/lernwege/alltag/abschluss/06-capstone" language: en image: "https://studiomeyer.academy/og/academy-og-share.png" publisher: "StudioMeyer Academy, https://studiomeyer.academy (llms.txt: https://studiomeyer.academy/llms.txt)" --- Level 2 was about Chain-of-Thought, system prompts, rolling context and tools. Now you turn those bricks into a system. Not a one-off, but something you can reuse two weeks from now. ## The task Pick a recurring task you spot in yourself. Real-world examples: - Every Monday assemble a weekly plan from calendar + emails + Notion todos. - For every LinkedIn comment decide answer/ignore/follow-up, with a draft reply. - Before every customer presentation, build a briefing (who is the customer, last contacts, open topics). - From a newsletter text, generate an Instagram post and a LinkedIn post, consistent in style. ## What the system has to do - **Input defined.** What exactly do you put in? An email? A calendar export? A customer name? - **Output defined.** Which format? Which fields? Which length? - **System prompt written.** In 4 to 8 sentences: the model's role, the rules that always apply, the output format. - **Chain-of-Thought where it makes sense.** For tasks where the model needs to give a reason (e.g. "why is this comment important enough to answer"). - **Tested with three different inputs.** One easy case, one tricky case, one edge case. Look at all three outputs and judge them. ## What you should produce A file `system.md` in your capstone folder. Containing: 1. What the system does (one sentence). 2. Full system prompt. 3. Input format (example). 4. Output format (example). 5. Three test runs with input and output. 6. Assessment: where did it work, where did it fail, what would you change. Plus the three outputs as separate files, exactly as they came out of the model, as evidence. ## What's strong The system is precise enough that someone else could pick it up 1:1. The system prompt is less than half a page but produces consistent outputs. You found at least one edge case where the model fails and documented how you handle it (fallback rule, or note "here I have to step in manually"). ## What's weak A system that only works for one example. A 3-page system prompt full of "please be professional" phrases. No test, no assessment. "Worked great" with no evidence. ## Bonus Build the whole thing as a small loop where you feed in a new input once a week and the system runs. One page of documentation that explains how someone else would use the system. That's the leap from "I wrote a prompt" to "I built a tool". ## Where it belongs Folder `portfolio/level-2-capstone/`. And if it turned out well: include it in your talent-pool profile. Companies want proof of "can work systematically", not "wrote one good prompt". ## Time 60 to 90 minutes. If it takes longer, the task is too complex. Start small. ## Wrap-up Take the [Level 2 checkpoint](https://studiomeyer.academy/checkpoint/2). Then Level 3, where automation enters (Zapier, n8n, webhooks). From there, "writing prompts" turns into "building processes". You can read everything freely. An account keeps your progress saved. [Sign in](https://studiomeyer.academy/en/login?next=%2Fen%2Flernwege%2Falltag%2Fabschluss%2F06-capstone) ## AI in everyday life [Back to learning path](https://studiomeyer.academy/en/lernwege/alltag) ## Your notes Sign in to save your own notes for this lesson. [Sign in](https://studiomeyer.academy/en/login?next=%2Fen%2Flernwege%2Falltag%2Fabschluss%2F06-capstone) ## Sitemap Public pages of this site as Markdown: [sitemap](https://studiomeyer.academy/sitemap.md). Curated entry point: [llms.txt](https://studiomeyer.academy/llms.txt), full text: [llms-full.txt](https://studiomeyer.academy/llms-full.txt).