Mockrill — Realtime AI Mock-Interview Voice Coach

Junior bootcamp graduates keep failing voice technical screens not from lack of knowledge but because they never rehearse speaking under pressure. Flashcards cannot hear filler words, follow-up panic, or pacing, and human coaching costs more than fifty dollars an hour. Mockrill is a realtime AI mock-interview voice coach built on AssemblyAI Path B. The browser streams microphone audio straight to AssemblyAI universal-3-5-pro over WebSocket, orchestrates questions and rubric scoring through the AssemblyAI LLM Gateway with structured tool calls, and speaks back with browser speech synthesis, so no second key or vendor is needed. Each answer is scored on structure, specificity, clarity, and relevance with word-level evidence, for example quoting kind of said three times at 07:42 with millisecond timestamps. Four spoken answers become a 3.1 overall scorecard, and the weakest answer is re-drilled in the same session with before-and-after deltas. The deployed demo runs on Vercel with a deterministic replay mode, and the business model pairs individual practice packs with per-cohort bootcamp licenses.