Meet Sunshine ☀️, the interactive AI meeting manager I built to end double-bookings

Built with n8n
, , , , , , ,
1

Meet Sunshine :sun:, the interactive AI meeting manager I built to end double-bookings.

Over the long holiday, I decided to push the limits of AI and workflow automation. But instead of building just another toy demo, I challenged myself to build something production-grade: Sunshine :sun:, a custom Meeting Manager System running entirely inside a real team’s LINE chat.

I wanted to tackle the messy, edge-case realities of daily operations and turn the dreaded “calendar ping-pong” into a frictionless, branded experience.

Here is the real business value Sunshine delivers:

  • Natural Language to Action: Type “Book a Team Sync-up with the CEO and Gorn on 19 May” → AI extracts details, translates nicknames to emails, and structures the data.

  • Smart Conflict Prevention: Cross-checks the Google Calendars of all required attendees in real-time. This completely eliminates the classic human error of double-booking that happens so often with manual calendar management.

  • Invisible Guardrails: Automatically enforces business rules using a code node (working hours only, no lunch break bookings) and even injects mandatory attendees if the user forgets them.

  • Interactive UI: Pushes beautiful, brand-aligned Confirmation and Cancellation Flex Messages directly in LINE.

  • Proactive Briefings: Generates a clean, automated daily schedule summary every evening so the team is prepped for tomorrow.

Built with:

  • n8n

  • GPT-4.1-mini / GPT-5-mini

  • Google Workspace APIs

  • LINE Messaging API.

My biggest takeaways from bridging the gap between prototype and production:

  1. Prompt engineering is the bottleneck: Getting reliable, strictly formatted structured output is much harder than configuring the tech stack itself.

  2. Model matching: Selecting the exact right model for each micro-job is a heavy, iterative process.

  3. Guardrails > Features: Anticipating human error and building silent safety nets matters way more than adding new bells and whistles.

  4. Know when to cut: I built a travel-time calculator but scrapped it. Reliability always beats feature count.

Ultimately, design must be driven by business impact, with technology acting as the engine that makes it real. The leap from a basic AI toy to a production-ready system is smaller than it seems. You just need the right architecture.

#AIAutomation #n8n #WorkflowAutomation #PromptEngineering #BusinessTransformation #Sunshine #MeetingManager

3

Hi Benjamin, thank you for your question!

The best practice is to have all nicknames different so that the data extractor node can pickup the correct email address. However, if any unavoidable matching of nicknames, I simply add their short last name e.g., James.a, James.o

All nicknames and email addresses mapping must be ready in lookup sheet before booking with the person. With exception for external domains where the system skips the mapping and get short name from email addresses instead.