Vibe Coding with Antigravity: How I Automate Meeting Notes in Real-Time.

Stop Listening to Meeting Recordings. Let AI Do It. (Built with Zero Coding)

If you work in a professional setting, you know the pain. The meeting ends, but the work isn’t done. You have a 60-minute audio recording, and now you have to listen to it all over again just to type up the minutes.

It’s time to stop.

I built a desktop application called Meeting Minutes AI. You simply drag and drop your audio file, and it uses Google’s Gemini AI to transcribe the audio and generate structured meeting minutes instantly.

The best part? I wrote zero lines of code to build this. I used a method I call “Vibe Coding.”

What is Meeting Minutes AI?

This is a professional desktop app designed for productivity. Here is what it does:

High-Quality Transcription: It uses the Gemini 2.5 Flash model to convert speech to text with incredible accuracy.
AI Summaries: It doesn’t just transcribe; it extracts Key Decisions, Action Items, and Structured Highlights.
Secure & Private: Your API key is stored locally on your machine. Your data is processed securely through your own Google AI Studio credentials.
Drag & Drop: Just drop an MP3, WAV, or WebM file, and the AI gets to work.

How I Built This (And How You Can Too)

I am a trader, not a software engineer. I don’t type code; I manage intent. This is the core of Vibe Coding. I act as the Product Manager and Architect, while an AI agent (like Google Antigravity, Cursor, or Windsurf) acts as my Senior Lead Developer.

I am sharing the exact “Vibe Prompts” I used to build this. You can open Google Antigravity right now and try this yourself.

Phase 1: The Blueprint

I started by telling the AI specifically what I wanted—a modern app using specific libraries.

My Prompt: “Create a modern desktop application using Vite, React, and Electron. Use a dark theme by default. Install the following libraries: lucide-react for icons, react-markdown for showing results, and @google/generative-ai for the AI connection.”

title: "The Vibe Coding Manifesto: Building Meeting Minutes AI"
slug: "vibe-coding-meeting-minutes"
date: "2026-02-13"
author: "Vibe Algo Lab"
tags:
  - "Vibe Coding"
  - "AI"
  - "Gemini"
  - "Electron"
  - "React"
description: >-
  Welcome to the future of software development. Vibe Coding is a paradigm shift where your primary tool isn't a keyboard, but your intent. This guide will show you how to build the "Meeting Minutes AI" app from scratch using only AI agentic coding assistants.

sections:
  - id: "what-is-vibe-coding"
    title: "💡 What is Vibe Coding?"
    content: >-
      Vibe Coding is about Agentic Programming. You act as the Product Manager and Architect, while the AI performs as the Senior Lead Developer. You don't write code; you communicate the "vibe"—the desired outcome, the aesthetic, and the functional logic.

  - id: "prerequisites"
    title: "🛠 Prerequisites"
    items:
      - name: "Node.js"
        description: "The engine that runs React and Electron."
      - name: "VS Code"
        description: "Your canvas."
      - name: "An Agentic AI"
        description: "(e.g., Antigravity, Claude Engineer, Cursor) that can natively read/write project files."
      - name: "Gemini API Key"
        description: "Get it at aistudio.google.com."

phases:
  - phase_number: 1
    title: "The Blueprint (Initialization)"
    goal: "Create a modern desktop app shell."
    prompt: |
      Create a modern desktop application using Vite, React, and Electron. Use a dark theme by default. Install the following libraries: lucide-react for icons, react-markdown for showing results, and @google/generative-ai for the AI connection. Set up a clean folder structure with electron/ for main processes and src/ for the UI.

  - phase_number: 2
    title: "The UI Canvas (Design)"
    goal: "Create a premium, state-of-the-art interface."
    prompt: |
      Design a premium 'Meeting Minutes AI' dashboard. The layout should be split:

      Left Panel: Audio player, file upload zone, and language selection.
      Right Panel: A scrollable AI Summary card.

      Apply a Glassmorphism style with subtle gradients, rounded corners, and smooth hover animations. Use a deep navy and electric blue color palette. Make sure it looks like a high-end SaaS product, not a simple MVP.

  - phase_number: 3
    title: "The Brain (Gemini AI Integration)"
    goal: "Make the app actually 'think.'"
    prompt: |
      Implement the AI logic using the Gemini 2.5 Flash model. Create a service that uploads an audio file to Gemini and performs two distinct steps:

      1. Transcription: Convert the whole audio to text.
      2. Summarization: Generate structured meeting minutes (Key points, Action Items, Decisions).

      Provide a callback for progress updates (e.g., 'Transcribing...', 'Summarizing...') so the user knows what's happening.

  - phase_number: 4
    title: "The Polishing (Iterative Debugging)"
    goal: "Fix the inevitable 'vibe-breaks' (errors)."
    prompt: |
      I'm getting an ENOENT error when I try to upload external files. I suspect it's because Electron can't access the file path directly. Can you modify the app so it first imports (copies) any selected file into a local Recordings/ folder within the app directory before sending it to Gemini? This will ensure the path is always stable.

  - phase_number: 5
    title: "The Delivery (Packaging)"
    goal: "Turn your code into a .exe file for users."
    prompt: |
      Configure electron-builder to package this app for Windows. I need two versions: a Setup Installer (.exe) and a Portable version that runs without installation. Make sure the 'Recordings' folder is handled correctly so files aren't lost when the app updates.

expert_tips:
  title: "🌟 Expert Vibe-Coding Tips"
  tips:
    - title: "Be Explicit about Aesthetics"
      content: >-
        Use words like "Premium," "Professional," "Sleek," and "Modern." If it looks basic, tell the AI: "The UI looks too generic. Add micro-interactions and better spacing to make it feel expensive."
    - title: "Iterate on Logic"
      content: >-
        If the summary is too short, say: "The AI summary is too brief. Change the prompt to Gemini to include a detailed breakdown of speaker intentions."
    - title: "Trust but Verify"
      content: >-
        Always ask the AI to run npm run lint or check the build status to ensure the code is production-ready.

footer: "Created with ❤️ for the Vibe Coding Community. Let's build the future together."

title: "The Vibe Coding Manifesto: Building Meeting Minutes AI"
slug: "vibe-coding-meeting-minutes"
date: "2026-02-13"
author: "Vibe Algo Lab"
tags:
  - "Vibe Coding"
  - "AI"
  - "Gemini"
  - "Electron"
  - "React"
description: >-
  Welcome to the future of software development. Vibe Coding is a paradigm shift where your primary tool isn't a keyboard, but your intent. This guide will show you how to build the "Meeting Minutes AI" app from scratch using only AI agentic coding assistants.

sections:
  - id: "what-is-vibe-coding"
    title: "💡 What is Vibe Coding?"
    content: >-
      Vibe Coding is about Agentic Programming. You act as the Product Manager and Architect, while the AI performs as the Senior Lead Developer. You don't write code; you communicate the "vibe"—the desired outcome, the aesthetic, and the functional logic.

  - id: "prerequisites"
    title: "🛠 Prerequisites"
    items:
      - name: "Node.js"
        description: "The engine that runs React and Electron."
      - name: "VS Code"
        description: "Your canvas."
      - name: "An Agentic AI"
        description: "(e.g., Antigravity, Claude Engineer, Cursor) that can natively read/write project files."
      - name: "Gemini API Key"
        description: "Get it at aistudio.google.com."

phases:
  - phase_number: 1
    title: "The Blueprint (Initialization)"
    goal: "Create a modern desktop app shell."
    prompt: |
      Create a modern desktop application using Vite, React, and Electron. Use a dark theme by default. Install the following libraries: lucide-react for icons, react-markdown for showing results, and @google/generative-ai for the AI connection. Set up a clean folder structure with electron/ for main processes and src/ for the UI.

  - phase_number: 2
    title: "The UI Canvas (Design)"
    goal: "Create a premium, state-of-the-art interface."
    prompt: |
      Design a premium 'Meeting Minutes AI' dashboard. The layout should be split:

      Left Panel: Audio player, file upload zone, and language selection.
      Right Panel: A scrollable AI Summary card.

      Apply a Glassmorphism style with subtle gradients, rounded corners, and smooth hover animations. Use a deep navy and electric blue color palette. Make sure it looks like a high-end SaaS product, not a simple MVP.

  - phase_number: 3
    title: "The Brain (Gemini AI Integration)"
    goal: "Make the app actually 'think.'"
    prompt: |
      Implement the AI logic using the Gemini 2.5 Flash model. Create a service that uploads an audio file to Gemini and performs two distinct steps:

      1. Transcription: Convert the whole audio to text.
      2. Summarization: Generate structured meeting minutes (Key points, Action Items, Decisions).

      Provide a callback for progress updates (e.g., 'Transcribing...', 'Summarizing...') so the user knows what's happening.

  - phase_number: 4
    title: "The Polishing (Iterative Debugging)"
    goal: "Fix the inevitable 'vibe-breaks' (errors)."
    prompt: |
      I'm getting an ENOENT error when I try to upload external files. I suspect it's because Electron can't access the file path directly. Can you modify the app so it first imports (copies) any selected file into a local Recordings/ folder within the app directory before sending it to Gemini? This will ensure the path is always stable.

  - phase_number: 5
    title: "The Delivery (Packaging)"
    goal: "Turn your code into a .exe file for users."
    prompt: |
      Configure electron-builder to package this app for Windows. I need two versions: a Setup Installer (.exe) and a Portable version that runs without installation. Make sure the 'Recordings' folder is handled correctly so files aren't lost when the app updates.

expert_tips:
  title: "🌟 Expert Vibe-Coding Tips"
  tips:
    - title: "Be Explicit about Aesthetics"
      content: >-
        Use words like "Premium," "Professional," "Sleek," and "Modern." If it looks basic, tell the AI: "The UI looks too generic. Add micro-interactions and better spacing to make it feel expensive."
    - title: "Iterate on Logic"
      content: >-
        If the summary is too short, say: "The AI summary is too brief. Change the prompt to Gemini to include a detailed breakdown of speaker intentions."
    - title: "Trust but Verify"
      content: >-
        Always ask the AI to run npm run lint or check the build status to ensure the code is production-ready.

footer: "Created with ❤️ for the Vibe Coding Community. Let's build the future together."

vibe-coding-guide.md Download

Phase 2: The UI Design

I didn’t want a boring gray box. I asked for a “Premium” feel.

My Prompt: “Design a premium ‘Meeting Minutes AI’ dashboard. The layout should be split: Left Panel for audio player and upload, Right Panel for the AI Summary. Apply a Glassmorphism style with subtle gradients, rounded corners, and smooth hover animations. Make sure it looks like a high-end SaaS product.”

Phase 3: The Brain (Gemini)

This is where the magic happens. I instructed the AI to connect the frontend to Google’s brain.

My Prompt: “Implement the AI logic using the Gemini 2.5 Flash model. Create a service that uploads an audio file to Gemini and performs two distinct steps: 1. Transcription (convert audio to text), 2. Summarization (generate structured meeting minutes). Provide a callback for progress updates.”

Download It for Free

If you want to study the code or just use the app to save time on your next meeting, I have uploaded everything to GitHub.

Download the App (.exe for Windows)
Get the Source Code

👉 Download at GitHub: vibealgolab/MeetingMinutes

Stop coding by hand. Start Vibe Coding.

“We respect your privacy. Your audio files and API keys are stored LOCALLY on your device and are processed via Google Gemini API solely for transcription. We do not store your data on our servers.”

Disclaimer

The software and code provided in this post are for educational and informational purposes only. They are provided “as is” without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and non-infringement. In no event shall the authors or copyright holders be liable for any claim, damages, or other liability, whether in an action of contract, tort, or otherwise, arising from, out of, or in connection with the software or the use or other dealings in the software.

This project is not affiliated with or endorsed by Google. Users are responsible for their own API usage charges and data privacy when using third-party AI services.

Post Views: 8