EchoPanel — Coordinated AI Interview Panel

Purpose: A voice-based interview platform where multiple distinct AI interviewer personas — Technical, Product/Business, Behavioural, Customer, Hiring Manager — conduct a single, real-time conversation with a candidate, coordinating with each other the way a real human panel would, rather than acting as one generic chatbot with a single voice.

The core problem it solves: Most AI interview tools ask a fixed list of questions and treat every answer the same way. A real panel doesn't work like that — a technical interviewer might accept a correct implementation, while a product interviewer independently pushes back because the candidate never explained the business impact. EchoPanel is built to reproduce that dynamic.

Goals

1. Adaptive questioning — questions are generated live based on what the candidate actually said, not a predefined script
2. Shared context across personas — every persona reads and writes to the same underlying "Context Graph" of claims, so a later interviewer can challenge something an earlier one accepted, without re-processing the whole transcript
3. Fair turn-taking — a "Turn Arbiter" decides who speaks next based on topic relevance and whether a claim has gone unchallenged, rather than a rigid round-robin
4. Difficulty that adapts to performance — a rolling per-topic competence score pushes questions from recall-level to applied to edge-case as the candidate demonstrates strength
5. Honesty checks — vague or contradictory answers are flagged automatically before being scored
6. Evidence-linked feedback — the final report ties every verdict back to a specific claim and transcript timestamp, not a vague summary
7. Real-time, interruptible voice — powered by Agora's Conversational AI Engine, so the interaction feels like a live conversation, not a form
8. Transparent AI disclosure — the candidate is always clearly told they're speaking with an AI panel, with explicit consent logged before the interview starts