📝

The Whiteboard Template

The whiteboard has a left column for text sections and a right area for your HLD diagram.

🏗️

System Components

Use rectangles for services, databases, caches, queues. Label every box — the parser reads the text inside to identify component types.

Keywords that are auto-detected:

servicedatabasedbsqlcacheredisqueuekafkarabbitmqgatewayload balancerlbworkercdndnspub-substorages3blob

➡️

Data Flow & Connections

Use arrows to connect components. Label them with what flows through, and number them for flow sequence.

💡 Tip: Numbered arrows help the AI understand your request flow order.

📋

Design Annotations

Add text boxes near components to explain design decisions. The AI reads these and gives you credit for thoughtful choices.

💡 Tip: Long text boxes (3+ lines) are treated as design rationale, not component labels.

📦

Service Clusters

To show multiple instances, draw a larger rectangle containing smaller copies. The parser auto-detects clusters and counts instances.

📊

Architecture Writeup

Fill in the left column sections with specifics. The more detail, the better the AI review.

💡 Tip: The more you fill in the left column, the better feedback you'll get. Empty sections = less to review.

💬

Explain Your Design

Click the Explain Designbutton in the top-right of the canvas to open the explanation panel. This is a free-text field — separate from your diagram — where you talk through your design as if you're in a live interview.

Cover your component choices, data flow reasoning, and key tradeoffs. The AI reads both your diagram andthis explanation together, so concrete reasoning here earns real credit — even if your diagram doesn't show every detail.

EXPLAIN YOUR DESIGN

0 / 2000 words

I chose Kafka over a direct DB write because the write volume at peak (≈ 50k msg/s) would saturate Postgres. Kafka lets me buffer and fan-out to both the timeline service and the notification worker without coupling them...

Redis stores the hot timeline (last 200 posts) with TTL=1h. For cold reads I fall back to Postgres with a covering index on (user_id, created_at DESC)...

💡 Tip: Think of this as your verbal walkthrough. The AI treats concrete mechanisms (“Kafka at 50k msg/s”) as real evidence. Vague claims (“it scales”) earn no credit.

✅

Pre-Submit Checklist

Every box has a descriptive label
Arrows connect from one box to another (not floating)
Key connections are labeled with what flows through them
At least FR and Assumptions are filled in
Design decisions are written as text near the relevant component
Explanation panel filled in — walk through your component choices, data flow, and key tradeoffs

📋 See a Real Example

View a completed system design with full AI review, responses, and re-evaluation.

View Example Design

Ready to put this into practice?

Start Drawing

Drawing Guide