KAVQEN: Turn Conversations into Completed Forms

Business intake forms push the work onto people: unclear fields, tedious typing, missing details, and staff follow-up to fill the gaps. Kavqen replaces the blank form with a guided voice conversation. An owner configures and publishes a workflow-backed form, then shares it through an email invitation or a time-limited access code. The participant signs in, starts a voice call, and talks the form through. Built on the AssemblyAI Voice Agent API, the agent asks one question at a time, clarifies answers, and confirms accuracy-sensitive details such as names, emails, and dates. Through schema-driven tool calls (save_form_field, submit_form, close_session), speech becomes application actions: fields update on screen as the participant speaks, answers can be corrected, and submission happens only on explicit request. The backend enforces the rules that matter: the Participant role is assigned server-side, form access requires ownership or a valid grant, invitations and codes expire, and submissions are validated with duplicate protection. Owners review structured submissions, reporting, and PDF exports, and can ground agents in their own knowledge documents via semantic search. Stack: Next.js, Node.js/Express gateway, Python/FastAPI, PostgreSQL with Prisma, Qdrant, and FastEmbed. The demo includes three workflows (Merchant Onboarding, Clinic Appointment Request, Customer Support Request) with fictional sample data, and currently supports English. Our first target is online-store support teams, where we plan pilots measuring completion rate, corrections, and staff follow-up time.