Building a Real-Time Voice AI Interview Coach for Job Seekers: Lessons in Latency, LLMs, and Speech Analytics

A developer built MockVora, a real-time voice AI mock-interview coach for job seekers, as a Hacktoberfest project. The system uses a three-tier pipeline designed to keep conversational latency under 1.5 seconds without relying on large cloud GPU infrastructure, targeting engineers who struggle to explain their reasoning aloud in English during placement interviews.

Every year during placement season, thousands of engineers face the same unspoken barrier: they know the code, but freeze when asked to explain their logic out loud in English. Reading technical solutions on GeeksforGeeks or LeetCode builds theoretical knowledge. But live interviews test speaking fluency, structure under pressure, and concise communication. For this Hacktoberfest Challenge, I set out to solve this exact problem by building MockVora https://mockvora.com/ —an end-to-end AI Voice Mock Interview coach that listens, adapts in real-time, and provides instant conversational feedback. Most traditional AI tools use text-based chat. However, real interviews are fundamentally verbal and auditory : To keep latency under 1.5 seconds without massive cloud GPU infrastructure, we designed a lightweight three-tier pipeline: