Preface

ML Sandbox is a hands-on learning platform for machine learning systems. Instead of just reading about concepts, you experiment with them — running real simulations, training actual models, and seeing how theory plays out in practice.

This platform is built around reputable ML sources, starting with Designing Machine Learning Systems by Chip Huyen. Each chapter pairs concise explanations with interactive sandboxes that make the concepts tangible. You don't just learn that batching trades latency for throughput — you drag a slider and watch it happen.

ML Sandbox is for engineers, data scientists, and students who want to build intuition for ML systems — not just models, but the infrastructure, tradeoffs, and human dynamics around them.

Read and interact. Each chapter has prose sections interspersed with interactive sandboxes. Read the context, then dive into the sandbox. The sandboxes are where the real learning happens.

Create a project. Projects are containers for your learning journey. When you create a project and set it as active, your sandbox runs and declared preferences are saved to it. Later chapters will adapt to your priorities.

Two types of sandboxes. Some sandboxes run server-side simulations streamed to your browser in real time (Pattern B — like the batching simulator). Others are fully client-side and respond instantly (Pattern A — like the stakeholder collision simulator). Both are designed to build intuition through doing.

Below are two sandboxes to get you started. The batching simulator lets you explore how request batching affects throughput and latency. The research-vs-production sandbox trains a real model and reveals how the same model behaves completely differently in research and production contexts.