Cloud Ops Chronicles

Just another devops lover

(C) 2026 Mathieu GOULIN

InkFlowAI: From Handwritten Notes to Structured Reports (Powered by AI)

Tue 20 January 2026

Do you take notes on your Kindle Scribe (or any e-ink tablet) only to end up with handwritten PDFs that sit in a digital graveyard? Here is how InkFlowAI automates the transformation of your scribbles into professional meeting minutes, diagrams included.

The Problem

Handwriting is fluid and natural for thinking, but terrible for sharing and archiving. Typing up notes takes time. Redrawing whiteboard sketches on a computer is tedious.

The Solution: InkFlowAI

A Serverless and Reactive system hosted on Google Cloud that acts as your personal secretary.

User Workflow

  1. You finish your meeting.
  2. You click “Share via Email” on your Kindle.
  3. 30 seconds later, you receive a reply email containing:
    • A structured summary (Markdown) of the meeting.
    • A list of action items and decisions.
    • Your handdrawn sketches converted into clean technical diagrams (Mermaid/PNG).

Technical Architecture (Under the Hood)

We recently migrated from a polling architecture to an event-driven (Push) architecture for real-time responsiveness.

Data Flow Diagram

image

Key Components

  1. Reactive Ingestion (Gmail + Pub/Sub): Instead of asking “Do I have mail?” every minute, our service sleeps. As soon as an email arrives, Gmail notifies Cloud Pub/Sub, which instantly wakes up our container on Cloud Run. This minimizes cost and latency.

  2. Multimodal Analysis (Gemini Pro Generative AI): The core of the system. We send the raw PDF (images of handwritten pages) directly to Gemini 3 Pro via Vertex AI. The system prompt is engineered to:

    • Decipher handwriting (Advanced OCR).
    • Understand context (brainstorming, decisions, TODOs).
    • Detect drawings and generate corresponding Mermaid.js code.

  3. Diagram Rendering (“NanoBanana”): If you drew a box pointing to a circle, Gemini translates it into code. InkFlowAI compiles this code into a sharp PNG image, ready to be embedded in the final report.

  4. Storage & Notification: Everything is archived on Cloud Storage (S3-compatible) for longevity. A formatted email is sent back to the sender.

Conclusion

InkFlowAI demonstrates how Generative AI (multimodal) can bridge the gap between the analog world (pen/paper) and the digital world (Jira/Wiki/Docs), automating the most boring part of work: cleaning up notes.

The code is Open Source and deployable on your own GCP project: https://gitlab.com/matgou/inkflowai