The interview is powered by Claude Haiku (Anthropic), a large language model selected for speed and cost-efficiency. A custom system prompt instructs the model to conduct a structured conversation across 20 barrier topic areas defined by DBE/ACDBE certification requirements.

As each topic is explored, Claude detects when sufficient detail has been gathered and emits a <<<SECTION_READY>>> marker — triggering the live narrative to build on-screen in real time.

Voice input uses the browser's built-in Web Speech API (SpeechRecognition), and text-to-speech playback uses SpeechSynthesis — no third-party audio service required. The microphone runs entirely in the browser; no audio is ever transmitted or stored.

The Anthropic API key is never exposed to the browser. All AI calls are routed through a Netlify serverless function acting as a secure proxy, with the key stored as an environment variable server-side.

When the interview is complete, the tool generates a formatted Microsoft Word document (.docx) in the browser using docx.js loaded from CDN. No server is involved — the file is assembled and downloaded entirely client-side.

The output follows the standard Personal Narrative of Social and Economic Disadvantage format: Times New Roman, structured section headings, a statement of purpose, barrier narratives, conclusion, and signature block — ready to submit for DBE/ACDBE recertification.