Multi Persona

Purpose: Let a recruiter send a candidate a link to a live, voice-based AI interview conducted by a panel of AI personas (e.g. technical, product, behavioral interviewers) instead of scheduling separate human screens.

How it works, at a high level:

• Setup — a recruiter configures the role, focus areas, which personas are active, and (per the in-progress plan) a time budget per persona and their own email for the report.
• Mic check — the candidate joins, tests their mic, then transitions into the live call.
• Conversation — a real-time voice session over Agora RTC/RTM, with one AI persona at a time conducting the interview, a live transcript, connection-health monitoring, and pipeline metrics (STT/LLM/TTS latency).
• Persona hand-off — the panel rotates through personas (manually today; the pending plan makes it timer-driven), carrying transcript context forward so the next persona doesn't repeat questions.
• Report — after the interview ends, a feedback report is generated (LLM-based with a deterministic heuristic fallback) and shown to the candidate; the pending plan adds emailing that report to the recruiter automatically.

Current goal / active work: get this reliably deployed to production on Vercel and finish the "timed panel rotation, recruiter email capture, and report delivery" feature so the whole flow — timed persona rotation, auto-end, and recruiter notification — works end-to-end without any manual clicking by the candidate.