Building a Sketchnote Artist with Gemini & Go ADK

Have you ever watched a technical talk and found yourself nodding along, only to realise an hour later that the key concepts have evaporated from your mind? That used to be me, until I discovered sketchnoting.

Years ago, I stumbled upon sketchnoting, the practice of creating visual summaries of content, and it transformed how I learn. As a visual learner, I remember not just what I read but where it sits on the page. By translating abstract concepts into spatial, visual notes, my retention improved dramatically. It was never about creating attractive pictures; it was about mapping information in a way my brain could navigate and recall.

But here is the problem: I cannot be everywhere at once, and I cannot sketch every video I watch.

So I asked myself: What if I could build an AI agent that watches videos for me and generates the sketchnotes I wish I had time to draw?

Enter the Sketchnote Artist

I am pleased to share Sketchnote Artist, an open-source project I built to automate this visual learning loop. Give it a YouTube video URL, and it transforms the content into a beautifully illustrated, hand-drawn-style sketchnote that captures the essence of what matters most.

Under the Hood: The Sequential Agent Workflow

This is not a wrapper around an API call. It demonstrates the Sequential Agent pattern using the Google Agent Development Kit (ADK) for Go*

I chose Go because its strong typing makes it ideal for building robust, production-ready agents, but mainly because it has fast cold-start times on Cloud Run. The architecture orchestrates two specialised agents working in sequence:

1. The Summarizer (Curator):

• Role Content Strategist
• Brain: Gemini 3 Flash. I chose Flash for its massive context window and speed. It can ingest entire video transcripts in one go and synthesise them into a structured ‘Visual Brief’, distilling the core thesis, key takeaways and compelling quotes.

2. The Artist:

• Role: Master Illustrator
• Tools: Gemini 3 Pro Image. This agent interprets the visual brief and orchestrates the image generation. The quality leap with Gemini 3 Pro Image is impressive: it handles text rendering and stylistic nuances, such as simulating the texture of alcohol markers on paper, with remarkable fidelity.

Flexible Workflow: CLI or Web

One of the strengths of the Go ADK is its versatility. I designed Sketchnote Artist to work however you prefer:

1. Local CLI: Perfect for developers who live in the terminal. Run it locally, pipe in a YouTube URL, and watch the agents work in real time.

2. Hosted Web UI: Prefer a friendlier interface? The project includes a server mode that deploys to Google Cloud Run, giving you a clean web UI where anyone can submit videos and view generated sketchnotes from any browser.

Seeing it in Action

The interaction is deliberately simple, regardless of how you run it. You provide a URL, and the system handles the complex orchestration behind the scenes.

CLI Example:

$ go run ./cmd/sketchnote console

User -> https://www.youtube.com/watch?v=dQw4w9WgXcQ

Behind the scenes, the Summariser processes the video, extracts the essential insights and hands off a baton, the Visual Brief, to the Artist. The Artist then ‘draws’ the final output, turning abstract concepts into concrete visuals.

The Result:

Instead of a wall of text or fragmented notes, you get a clean, visual artefact like this:

Example Sketchnote

Why This Matters

As a Google Developer Expert, I have seen countless AI demos. But building this one felt different. It was a convergence of capabilities that made something I have wanted for years practical:

• Gemini 3 Flash makes the “reasoning” phase fast enough to feel seamless
• Building a Sketchnote Artist with Gemini & Go makes the “creation” phase high-quality enough to be genuinely useful
• Go ADK makes the plumbing robust enough to be maintainable and production-ready

This is not about impressive technology alone. It solves a problem in how I consume and retain knowledge.

Try It Yourself

This project has been a pleasure to build, not because the tech stack is exciting, but because it improves my learning workflow.

I would be delighted if you gave it a spin. Whether you want to generate sketchnotes for your own learning or dive into the code to see how a multi-agent system is structured in Go, the repository is ready for you.

Clone it, run it locally, and let AI sketch your next learning session.

https://github.com/danielvaughan/sketchnote-artist

Live Demo

I also have a version deployed at

https://sketchnotes.danielvaughan.com

Please note that access is restricted to manage costs. If you’d like to try the live version, please email your Google email address to sketchnotes@danielvaughan.com, and I’ll be happy to onboard you.

Happy coding (and sketching)!