[grid mode=carousel]
[/grid]
5 years ago, I built a graph algorithm visualizer in about 30 days. It got great feedback on LinkedIn, but over time I felt it no longer reflected how I actually wanted it to look, feel, or work. So I decided to rebuild it properly.
Why now?
Towards the end of 2025, the hype around Claude Code on X was impossible to ignore. I decided to lean into it and see how far I could push a serious rebuild using AI-assisted development.
Tools & Stack
-
Claude Code Max with heavy use of Plan Mode
-
Migrated from CRA to React 19 with Vite
-
Tailwind v4
-
Zustand for state management
-
A mix of Agent skills like Claude’s frontend-design, Vercel’s web-design-guidelines and more.
Process & Workflow
-
I started by migrating the old CRA codebase to a modern React 19 and Vite setup. Claude handled this surprisingly fast with minimal issues.
-
Once feature development picked up, I hit Claude’s Pro tier limits. To avoid breaking flow, I upgraded to the Max plan, which was worth it for uninterrupted iteration.
-
Early implementations worked, but the generated code relied heavily on window state and CustomEvent dispatchers. I paused, switched to Plan Mode, and refactored everything around Zustand.
-
I iterated heavily on UX and landed on an Excalidraw-inspired interaction model. Creating nodes, dragging them, and connecting edges started to feel intuitive and natural.
-
Undo and redo took multiple iterations. Eventually, I abstracted it cleanly into a higher-order function inside the Zustand store, which simplified everything.
-
For UI design, I worked closely with Claude to tune shadows and depth to get a subtle skeuomorphic feel. After a few rounds, I stripped redundancy and enforced a strict component-first design system.
-
Finally, I ran accessibility and performance checks by using agent skills.
Key Learnings
-
Plan Mode plus constant refactoring was the real superpower.
-
AI-generated code works best when you aggressively review, restructure, and raise abstraction levels.
-
AI is incredible for momentum, but design taste and architectural decisions still need a human in the loop.
Outcome
I worked on this for 7 days and ended up with far more features than I originally planned. If I had built this from scratch without AI, it would have easily taken 2 months, mostly due to documentation, experimentation, and dead ends.
Links
App: https://graphisual.vercel.app Code: https://github.com/lakbychance/graphisual
Would love any feedback, especially on UX, architecture, or performance.