During the Fall 2025 semester, I took Cornell’s Info 5940 course about AI Chatbots, RAG, AI Agent. The course was taught by Prof. Ayham Boucher and focused on understanding the systems behind Large Language Models and building Chatbots, Agents, and Retrieval-Augmented Generation (RAG) Applications. It was engaging and eye-opening. Every session we discussed the current AI events of the day, we had captivating guest speakers from leading companies in the field (OpenAI, Google, Boomi), and we were guided through hands-on development of AI focused applications. The final group project involved solving a real-world problem by leveraging what we learned during the semester. Below is a summary of my group’s project.

The hospitality industry faces a persistent challenge: how do you quickly and effectively train new front desk staff on your property’s unique language, culture, and service standards without overwhelming guests or creating stressful real-world learning experiences? My team tackled this problem head-on, developing an innovative Hotel Front Desk Training Agent system that uses AI to create realistic, controlled training environments for hospitality workers.

The Problem

Traditional hotel training often throws new employees directly into guest-facing situations, where mistakes can impact customer experience. Each property has its own:

• Service standards and protocols
• Specific language and terminology
• Cultural expectations
• Unique policies and procedures

Standardizing this onboarding process while maintaining quality has been an ongoing challenge for hospitality managers.

Our Solution: Multi-Agent AI Training Environment

We created a multi-agent environment where new hires can gain experience handling stressful guest interactions in a controlled digital space. The system uses:

• RAG (Retrieval-Augmented Generation) applications to ensure accurate, property-specific responses
• Multiple interacting AI agents that simulate realistic scenarios
• Real hotel data provided by our industry partner, The Statler Hotel

The key innovation is that trainees can practice difficult situations—angry guests, complex requests, policy conflicts—without any risk to actual customer relationships.

Real-World Partnership

This wasn’t just an academic exercise. We partnered with The Statler Hotel and their Director of Rooms, Sam Everett, who provided actual:

• Training materials
• Service standards
• Property-specific information
• Policy documentation

This real-world data was structured into a YAML database that powers our AI agents, ensuring they respond with authentic, hotel-specific information.

The Human Element Remains Central

Importantly, this system is designed as a supplement, not a replacement for human training. As we emphasized in our presentation: “Human connections are the essence of hospitality.”

The goal isn’t to replace the experience gained through real guest interactions, but to give new employees a foundation of confidence and competence before they engage with actual guests.

Technical Implementation

Our system is built around three specialized AI agents working together to create an immersive, responsive training environment:

The Three-Agent Architecture

1. The Guest Agent: Simulates realistic hotel guest scenarios and interactions. This agent creates challenging customer situations—lost reservations, noise complaints, maintenance issues—and responds dynamically to the trainee’s approach. If the trainee is empathetic, the guest calms down; if the trainee is rude, the guest becomes more frustrated.

2. The Coaching Agent: Provides real-time, private feedback to trainees during their interaction with the guest. It analyzes responses against hotel policies and offers instant guidance in the sidebar, helping trainees learn proper phrasing, service standards, and brand-appropriate responses.

3. The Report Agent: After each session, this agent performs a comprehensive audit of the entire interaction. It generates detailed performance reports with executive summaries, performance highlights, development opportunities, and specific training recommendations for both the trainee and their manager.

Three-Phase Training Process

• Phase 1 – Initialization: When a trainee clicks “Start,” the system loads hotel-specific training materials from a YAML database into memory, ensuring all agents operate according to the property’s exact policies rather than general knowledge.
• Phase 2 – Interactive Loop: The core training experience happens entirely within a web interface. Trainees engage in realistic conversations with the Guest Agent while receiving real-time coaching feedback, creating a fully responsive simulation environment.
• Phase 3 – Reporting: The system doesn’t just end—it analyzes the full conversation history to provide actionable insights, scoring trainees against evaluation criteria and offering concrete next steps for improvement.

The Power of Consistency

The entire system is grounded in a single source of truth: the YAML training database containing actual hotel policies and procedures. This ensures every AI agent operates according to specific brand standards, creating consistent, scalable role-play training that’s deeply aligned with the property’s service philosophy.

Looking Forward

This project demonstrates how emerging AI technologies can address real industry challenges while preserving what makes hospitality special—genuine human connection. By giving new staff the tools and confidence they need upfront, we can improve both employee success and guest satisfaction.

This project was completed as part of Cornell’s Info 5940 course by Dominic Chatas, Will Olson, Sanjeev Ragunathan, and Laurence Yang.