Casa Noor, AI-Powered Hospitality Platform
Casa Noor — AI-powered hospitality and travel platform
3,000+
Bookings per month
−35%
Support tickets
60%
Queries resolved by AI
< 2s
First-token latency
Overview
Repetitive booking queries were overwhelming the support team. We built an AI assistant into the web platform that now handles 60% of queries autonomously and drives 3,000+ monthly bookings.
The challenge
What we were up against
Casa Noor's web platform routed all booking queries through live chat to human agents. Most were repetitive tier-1 questions that didn't need a person: availability checks, cancellation policies, room type comparisons. The support cost per booking was climbing, and response times during peak seasons pushed guests to competitor platforms. The team needed an AI assistant that could handle routine queries autonomously, hand off complex cases cleanly to a human agent, and perform reliably across devices and connection speeds.
The approach
Engineering decisions
Designed the conversation flow architecture first: query classification, confidence thresholds for autonomous response vs. human handoff, and explicit fallback handling to prevent hallucinated room availability or pricing.
Integrated the server-side LLM API with streaming response support using React Server Components and edge functions, rendering tokens progressively to reduce perceived wait time from 3s to under 2s first-token.
Built a resilient message queue so conversations survive network interruptions and page refreshes, with automatic retry and a clear connection status indicator for guests.
Implemented graceful degradation: if the LLM API is unavailable or response confidence falls below threshold, the assistant routes to a human agent with full conversation context pre-loaded so guests never have to repeat themselves.
Added structured logging to track query type distribution, model confidence, handoff rate, and resolution rate, giving the product team data to tune prompts and measure ROI post-launch.
The outcome
Results
The AI assistant now drives over 3,000 bookings per month and handles 60% of all booking queries autonomously. Tier-1 support ticket volume dropped 35%, significantly reducing operational costs. The human handoff rate stabilized at 18%, well within the product team's target. First-token latency sits under 2 seconds across all device types.
3,000+
Bookings per month
−35%
Support tickets
60%
Queries resolved by AI
< 2s
First-token latency
Tech stack