
A US B1/B2 visa interview lasts about two minutes, and the officer already has your filed DS-160 on screen. It is not a conversation but a cross-examination of that form, under a legal presumption that you intend to immigrate (INA 214(b)) and a burden that sits on you to rebut it. Officer Parker rebuilds that situation as a voice agent. 1. You file a short DS-160-style form: purpose, occupation, who pays, trip length, US relatives, prior travel, return plan, home ties. Then you speak, and a simulated consular officer interviews you live, one short question at a time, brisk and never coaching. Two-minute window mode asks three to five questions and ends, like the real thing. 2. The officer holds your form. When what you SAY diverges from what you FILED, it interrupts: "You said your company is paying. You filed that you are retired. Which is it?" 3. You get a report: every contradiction, quoted from both sides; scores across six dimensions; a deterministic filler count; a numbered fix list. A job interview asks whether you are the best candidate. A visa interview asks whether you rebutted a presumption, against a document you already signed. Contradicting your own form is not a weak answer there, it is a refusal ground, and a resume-based interviewer cannot check for it. ASSEMBLYAI: one Voice Agent API WebSocket carries the whole live loop - streaming speech-to-text, turn detection, barge-in, LLM routing for the officer persona, and its voice. The browser never sees the API key; a server route mints a 300-second single-use token. Scoring runs through the LLM Gateway. HONESTY: no rewrite it suggests may introduce a fact you did not state. Where an answer cannot be improved without inventing one, it says so. Coaching someone to lie at a US embassy window gets them barred for life. LIMITS: never validated against real visa outcomes; it does not predict yours. A practice tool, not legal advice, not affiliated with the State Department.
30 Sep 2026