Interview Preparation: 150+ Questions & Answers

Category 1: Edge AI & Computer Vision

1. Why run MediaPipe in WebAssembly instead of a backend server?
   • Running inference locally eliminates network round-trip latency (dropping response times from 200ms to 15ms), eliminates expensive cloud GPU server hosting costs, and ensures 100% user privacy since camera frames never leave the client's device.
2. How does the MediaPipe Hand Landmark model work?
   • It employs a two-step pipeline. A palm detector first identifies hand presence across the frame. Once detected, a hand landmark model predicts the 3D coordinates of 21 distinct joints (knuckles, fingertips, wrist).
3. What is the purpose of the GPU delegate?
   • By default, WebAssembly executes on the CPU. Specifying delegate: "GPU" directs MediaPipe to compile shader programs using WebGL, delegating heavy matrix convolutions to the GPU for smooth 60fps throughput.

Category 2: Performance & Frontend Engineering

4. How do you avoid freezing the React UI during continuous video prediction?
   • By orchestrating predictions inside requestAnimationFrame. Instead of running an unthrottled while loop, each frame prediction is synchronized with the browser's native screen refresh cycle.
5. How did you prevent gesture flapping?
   • By combining a strict confidence threshold ($> 0.75$) with a 1500ms time-decay debounce window stored in a React ref. Once a command executes, subsequent commands are suppressed until the timer expires.