Building Three21 pushed us across the full technical spectrum 3D mathematics, graph algorithms, AI orchestration, real-time streaming, deployment constraints, and global accessibility. One of the hardest problems was creating an Iron-Man-style 3D hierarchical breakdown system that felt precise and stable at 60 FPS. This required BFS-based layer detection, first-principles vector math for clean explosion directions, memoization to eliminate per-frame O(n) costs, and a strict LIFO stack for perfect reassembly. On the platform side, our SSE-based /api/chat route initially failed on Vercel due to the Edge Runtime’s lack of Node buffer support; explicitly forcing the Node.js runtime resolved multimodal crashes instantly. Maintaining uninterrupted AI streaming alongside background research tools surfaced an unexpected UI-level bug Tailwind z-index layering caused component re-mounts that broke streams. Re-architecting the UI to persist surfaces and backing chat state with IndexedDB resulted in a stable, offline-friendly assistant. Rendering posed another challenge: Three.js offered little guidance for hybrid WebGPU/WebGL setups, so we implemented our own detection and fallback system, gaining an 8–14% FPS boost where WebGPU was available. On the AI side, research hallucinations were reduced by tightening prompts, lowering temperature, sanitizing queries, and limiting tool loops. Finally, global accessibility demanded accurate translation into 14 languages; Lingo.dev enabled compiler-level UI translation, correct technical terminology, and RTL support, expanding our reach to over 5 billion users. Together, these challenges reinforced a core lesson: first-principles thinking, classic DSA, careful UI architecture, and disciplined prompt design are what transformed Three21 into a robust, production-grade platform.