flow works flawlessly, the same agent should also help with booking spa sessions or late-checkout, share up-to-date tips on local attractions, and generally answer every routine question a front-desk concierge receives. Core requirements • Conversational fluency in English, Spanish, and Chinese, with automatic language detection so guests are never forced through a menu. • Real-time integration with Opera PMS for live room status, billing, and service ticket creation. The integration must read and write data securely via Opera’s APIs. I’d like the architecture kept modular so we can later extend it to systems such as RoomRaccoon or Cloudbeds without a full rebuild. • 24/7 availability with natural, human-sounding voices, minimal latency, and graceful fallback to a human operator if confidence scores drop below an agreed threshold. • Centralised analytics dashboard logging every interaction, categorised by intent (room service, amenities, FAQs, etc.) for continuous training and service-quality tracking. • Simple configuration panel where our staff can update menu items, service fees, or local-attraction snippets without touching code. Deliverables 1. Deployed voice agent (cloud or on-prem) connected to Opera PMS and reachable by SIP/VoIP number we provide. 2. Source code with clear README and environment files. 3. API and webhook documentation covering authentication, payload examples, and error handling. 4. One-hour remote walk-through for our IT staff plus test scripts for acceptance. Acceptance criteria • A guest calling from Room 101 can order “a Caesar salad and two sparkling waters” and see the ticket appear in Opera within 10 seconds. • Language switch verified by starting the same call in Spanish and Chinese. • Average intent-recognition confidence ≥ 90 % across 50 sample questions. • All PII encrypted in transit and at rest, meeting GDPR standards. If you have prior experience with Dialogflow CX, Amazon Lex, or similar NLU engines tied into hospitality platforms, please highlight it when you respond.