Gauntlet AI (The First 24 Hours)

I Built a Real-Time Collaborative Whiteboard From Scratch in Week One at Gauntlet AI

My first project at Gauntlet AI was to build a production-ready Miro clone — a real-time collaborative whiteboard with authentication, live cursors, shared boards, and canvas tools. From zero to deployed MVP.

Here's what I shipped and how I built it.

What the App Does

The MVP is a fully functional collaborative whiteboard. Users authenticate, land on a dashboard of their boards, and can create as many workspaces as they want. Inside each board, there's a full suite of canvas tools — sticky notes, shapes with customizable colors, text, titles — all rendered on a live canvas.

The real feature is real-time collaboration. Multiple users see each other's changes as they happen, with avatar indicators showing who's on the board. Share a link, and anyone can authenticate and join the session instantly.

The Tech Stack

I went with three core tools that let me move fast without sacrificing production quality:

Vercel for deployment and hosting — instant previews on every push, zero config.

Supabase for database and auth — real-time subscriptions out of the box, which made the collaboration layer significantly easier to wire up.

Canvas API for the whiteboard rendering — direct control over the drawing surface without the overhead of a heavy framework.

How I Built It: Planning Before Prompting

The biggest takeaway from this build wasn't a technical trick — it was a process discipline.

Before writing a single line of code, I created a structured development plan. Shout out to Ash from the cohort, whose breakdown on optimizing Cursor development reinforced something I already believed but needed to be more intentional about: the quality of your AI-assisted output is directly proportional to the quality of your planning inputs.

My project file structure went beyond the code itself. It included:

A custom PRD built from the project requirements provided during onboarding — not a generic template, but a document tailored to this specific build with clear scope, constraints, and success criteria.

A system design document outlining the architecture before touching an editor.

A tickets.md file breaking the entire PRD into discrete, sequenced tickets — each one representing a shippable unit of work.

A test-driven development methodology where every ticket started with defining what "done" looks like before building toward it.

The Cursor Workflow That Kept Me Fast

Every ticket got its own plan inside Cursor using the planning feature. I'd scope the work, let the agent execute against it, and review the output against my test criteria.

The key habit: after completing each ticket, I started a fresh agent session. This kept the context window clean and prevented the compounding confusion that happens when an AI assistant is carrying too much stale context. Small discipline, massive impact on output quality.

The Hard Part

Persistence was the main issue I ran into. Intermittently, refreshing the app would lose board state — the kind of bug that's invisible until it isn't. It was a classic real-time sync problem: making sure that what's in memory, what's in the database, and what's on screen all agree, all the time. It took focused debugging, but it got solved.

Why This Matters

This wasn't a tutorial project. It was a production deployment built under time pressure with real architectural decisions — auth flows, real-time data sync, canvas rendering performance, multi-user state management.

Week one at Gauntlet AI, and I'm already shipping things I would have scoped as multi-sprint efforts in my previous role. The difference isn't just AI tooling — it's the combination of rigorous planning and AI-assisted execution. The tooling amplifies your thinking, but only if your thinking is structured.

Next up: integrating LangChain into the platform. That's where it gets interesting.

The walkthrough video is at the top of this post. Try the app here: https://lnkd.in/gfC4FsZX

---

Originally published as a LinkedIn article and post.