It is a platform for learning both Classical System Design (load balancers, databases, caches, queues) and Modern AI System Design (LLMs, RAG, agents, MCP, guardrails) — the two tracks every engineer needs for senior-level interviews and real-world architecture work.
The learning loop has three steps:
- Learn — Animated lessons with real company examples, trade-off breakdowns, and interview questions for every concept
- Build — Drag-and-drop canvas to design architectures. Guided mode walks beginners through step-by-step with hints. Free mode for experienced designers
- Review — AI reviewer scores your design across 8 categories, identifies bottlenecks, asks follow-up interview questions, and shows you the reference architecture
- Classical Systems — 10 lessons: Client–Server, Load Balancer, API Gateway, Microservices, Databases, Caching, Message Queues, CDN, Scalability, Reliability
- AI & MCP Systems — 15 lessons: LLM APIs, Prompt Engineering, Embeddings, Vector Databases, RAG, AI Agents, Tool Calling, MCP, Multi-Agent Systems, AI Memory, Guardrails, AI Evaluation, AI Observability, AI Security, and more
Each lesson includes:
- Animated architecture diagram (unique to each concept)
- Key insight
- Real company examples (how Netflix, Uber, Slack actually use this)
- Advantages and disadvantages
- Interview questions
Before tackling challenges independently, beginners get step-by-step guidance. Each guided step presents a multiple-choice question about what component to place next. Correct answers auto-place the component on the canvas with an explanation of why. Wrong answers give a hint without revealing the answer. Completing all steps unlocks free-build mode.
- Drag-and-drop canvas (React Flow)
- 24 components across 7 categories: Client, Network, Compute, Storage, Messaging, AI/ML, Ops
- AI-specific components: LLM API, Embedding Model, Vector Database, MCP Server, AI Agent, Memory Store, Guardrails, Monitoring
- Interview tips panel per challenge
- Written reasoning textarea
Classical:
| Challenge | Difficulty |
|---|---|
| Design Uber | Hard |
| Design WhatsApp | Hard |
| Design Instagram | Medium |
| Design YouTube | Hard |
| Design a Payment System | Medium |
AI System Design:
| Challenge | Difficulty |
|---|---|
| Design an AI Coding Assistant | Medium |
| Design an AI Customer Support Agent | Easy |
| Design a Research Agent using MCP | Hard |
| Design a Document Q&A System | Easy |
Every challenge includes: problem statement, functional requirements, non-functional requirements, expected scale, constraints, interview tips, guided build steps, and a reference solution.
Submit your design and get back:
- Overall score (0–100) with a hiring verdict
- Category scores: Functionality, Scalability, Database Design, Performance, Security, Reliability, Cost Awareness, Design Reasoning
- Identified strengths and bottlenecks
- Concrete improvement suggestions
- Follow-up interview questions a real interviewer would ask
After reviewing, click "View Reference Solution" to see the expert architecture on an interactive React Flow canvas. Click any component to learn exactly why it exists in the design. Includes architecture breakdown, key decisions with reasoning, and a component checklist.
Full dark theme with smooth transitions. Respects system preference on first visit. Persists across sessions. Toggle with the button in the bottom-right corner.
| Layer | Tech |
|---|---|
| Framework | React 18 + Vite 6 |
| Routing | React Router v6 |
| Canvas | React Flow (@xyflow/react) |
| Styling | Tailwind CSS v4 |
| Animations | Framer Motion |
| Icons | Lucide React |
| AI | Groq API (llama-3.3-70b-versatile) |
| Theme | CSS custom properties + .dark class |
No backend required. The AI review calls the Groq API directly from the browser.
- Node.js 18+
- npm 9+
git clone https://github.com/your-username/system-design-arena.git
cd system-design-arena
npm installnpm run devOpen http://localhost:5173.
Create a .env file in the project root:
VITE_GROQ_API_KEY=gsk_your_key_here
Get a free key at console.groq.com. The app runs fully in demo mode without one — you just get a realistic mock scorecard instead of a live AI review. Restart the dev server after adding the key.
When you submit a design, the workspace serializes your diagram into a structured JSON blueprint:
{
"challengeId": "uber",
"challengeTitle": "Design Uber",
"nodes": [{ "id": "n1", "type": "network", "label": "Load Balancer" }],
"edges": [{ "source": "n1", "target": "n2" }],
"reasoning": "I added a load balancer to..."
}This blueprint — along with your written reasoning — is converted into a system design interview prompt and sent to llama-3.3-70b-versatile via the Groq API. The model is instructed to respond with structured JSON only, which is parsed into the scorecard displayed on the review page.
The model evaluates: what components you used, how they're connected, whether critical components are missing, and whether your written reasoning explains the trade-offs.
Built to teach engineers not just what components to use, but why they exist and how they work together.