Build a real pipeline | Master AI Automation in 4 hours Master AI Automation in 4 hours Course About Ayush Modules Sample chapter Toolbox The Microcap Minute Classroom / Module 06: Webhooks & Automation / Chapter 3 Build a real pipeline Watch first, then read. Same lesson, your pace. What you will learn – A four-step automation you’ll actually keep – Mapping data between steps – Testing like an engineer, not a hopeful The brief Real scenario, real value: a study-group feedback form . Every submission should be instantly (1) understood by AI, sentiment + key theme, (2) logged to a sheet, (3) flagged to your phone if someone’s struggling. Manually that’s five minutes per response; automated it’s nine seconds forever. Assembling in n8n Four nodes, one wire each: Node 1, Webhook trigger. New workflow → Webhook node. Copy its test URL. Node 2, Google Form entry. For the course version, simulate submissions with curl at the test URL: curl -X POST -H “Content-Type: application/json” -d ‘{“name”: “Priya”, “topic”: “MCP servers”, “rating”: 2, “comment”: “Chapter 3 config confused me completely”}’ (Production alternative: n8n’s Google Forms/Sheets triggers do this natively once you OAuth them.) Node 3, The AI step. Add an OpenAI node or HTTP Request node pointed at OpenRouter (your :free models from Module 4). Prompt, R-C-T-F, naturally: You are a teaching assistant. Given student feedback JSON, return ONLY valid JSON: {“sentiment”:”positive|neutral|concerned”, “theme”:””, “action”:””} Map the payload fields into it ( etc. via drag-and-drop mapping). Structured-in, structured-out: Module 4 Chapter 2 earning its keep. Node 4, Log + alarm. Append the original feedback + AI verdict as a row in Google Sheets. Then an IF node: sentiment = “concerned”? → send a Telegram/Slack/Discord message (“⚠️ Priya stuck on Ch3”). Else: nothing. That branch is your error path thinking from Chapter 2, applied. Activate the workflow. Fire five varied curl payloads. Watch rows appear and one alert arrive. Test like you mean it Amateurs test with happy paths only. You: Send a rating of 5 and of 1, both branches fire? Send an empty comment, does the AI node choke or cope? Send malformed JSON, does the flow die loudly or log gracefully? Every failure found now is a midnight emergency avoided later. Try it yourself Build all four nodes; run the five-payload test suite above. Then personalise: change the AI’s output schema to add “difficulty”:”easy|medium|hard” and update the sheet columns to match. Screenshot the final canvas into learn/pipeline.png . Keep this workflow alive, Chapter 5 turns it always-on. Key takeaways – Trigger → enrich with AI → store → conditional alert is the canonical automation shape. – Mapping drags earlier steps’ data into later prompts; JSON-in/JSON-out keeps machines happy. – Branch on AI verdicts, automations that decide beat automations that just record. – Test sad paths deliberately; hope is not a strategy. Download the exercise sheet (PDF) Module workbook (PDF) ← Prev: No-code automation Next: Roll your own receiver → Classroom / Module 06: Webhooks & Automation / Chapter 3 Build a real pipeline What you will learn – A four-step automation you’ll actually keep – Mapping data between steps – Testing like an engineer, not a hopeful The brief Real scenario, real value: a study-group feedback form . Every submission should be instantly (1) understood by AI, sentiment + key theme, (2) logged to a sheet, (3) flagged to your phone if someone’s struggling. Manually that’s five minutes per response; automated it’s nine seconds forever. Assembling in n8n Four nodes, one wire each: Node 1, Webhook trigger. New workflow → Webhook node. Copy its test URL. Node 2, Google Form entry. For the course version, simulate submissions with curl at the test URL: curl -X POST -H “Content-Type: application/json” -d ‘{“name”: “Priya”, “topic”: “MCP servers”, “rating”: 2, “comment”
Build a real pipeline
Written by
in