Voice Triage Agent

THE PROBLEM Support calls are messy. Callers pause, restart, go off-topic, and often mention the key detail last ("I have a client demo in an hour"). Agents then retype everything into a ticket and guess the category and priority. THE SOLUTION Voice Triage Agent automates that first step. The caller describes the problem out loud; the agent listens in real time, asks one follow-up question at a time by voice, and produces one structured ticket with category, priority, summary, and full transcript. HOW IT WORKS 1. Speech streams live to AssemblyAI Universal Streaming, which returns formatted turns in real time. 2. Each turn, with the full conversation, goes to Google Gemini, which updates one ticket and decides the agent's next reply. 3. The reply is spoken back. The mic is muted while the agent speaks, so it never hears itself. 4. When the call ends, the ticket is saved as JSON with the transcript. HOW WE USE ASSEMBLYAI - Universal Streaming for low-latency live transcription - Tuned turn detection so mid-sentence pauses don't cut callers off - Key-term prompting for support vocabulary ("log in", "billing", "client demo") WHAT MAKES IT DIFFERENT - One call = one ticket: fragments are merged and transcription errors corrected from context - Context-aware priority: deadlines raise urgency automatically - Stays on task: off-topic requests are politely redirected - Always-valid output: strict JSON schema BUSINESS VALUE Faster first response, consistent categories, and correct priorities with no retyping. Fits helpdesks, IT service desks, and SaaS support; tickets are ready for Jira, Zendesk, or Freshdesk. NEXT STEPS Phone-line integration (Twilio), direct helpdesk integration, a live dashboard, and barge-in. TECH STACK Python 3.12, AssemblyAI Universal Streaming, Google Gemini, PyAudio, Windows speech voices.