Agentic AI HotelBot

AI-powered virtual concierge for luxury hotels using GPT, LangGraph agents, LlamaIndex, FastAPI, Redis, and Neon PostgreSQL. Supports room booking, FAQ search, and real-time availability. Deployed via Docker on Hugging Face and AWS ECS.

Agentic AI-Powered Virtual Concierge for Luxury Hotels.
This diagram shows the architecture of a multi-agent hotel concierge system powered by GPT-4 and LangGraph. A Supervised Agent coordinates a RAG agent (FAQ retrieval via LlamaIndex) and an SQL agent (room availability via PostgreSQL). FastAPI connects the backend to a Streamlit UI, with Redis used for storing chat history.

Blue Horizon AI Concierge

AI-Powered Virtual Concierge for Luxury Hotels

AI-Powered Guest Interaction & Real-Time Booking

The Blue Horizon AI Concierge is a full-stack virtual assistant system that brings natural language interaction to luxury hospitality. It interprets guest queries using LLMs, enables real-time room bookings, and provides intelligent, context-aware responses for FAQs and hotel services.

1. Introduction

Luxury hotels often lack scalable, conversational systems to manage dynamic guest inquiries, real-time booking, and service personalization. Blue Horizon solves this by integrating state-of-the-art LLMs, semantic search, and multi-agent coordination to build a real-time concierge application with seamless frontend-backend communication.

2. Key Questions Addressed

  • How can we use natural language understanding for hotel bookings?
  • Can LLMs interpret unstructured guest queries and convert them into database actions?
  • How can a dynamic FAQ system improve guest experience?
  • How do we deploy a scalable concierge application with minimal latency?

3. The Problem

  • Traditional booking systems are rigid and rule-based.
  • Guests need to rephrase queries to match the system.
  • FAQs are static and non-personalized.
  • Hotel systems don’t utilize real-time data pipelines or multi-agent intelligence.

4. The Importance

  • Increases guest satisfaction through intelligent, real-time responses.
  • Automates concierge services, reducing staff workload.
  • Enables dynamic pricing and booking via real-time database queries.
  • Provides flexible deployments through Docker, Hugging Face, and AWS.

5. The Solution

5.1 Conversational Intelligence

  • Uses OpenAI GPT for query parsing.
  • Implements NL-to-SQL conversion for booking intent via a dedicated agent.

5.2 Multi-Agent Orchestration with LangGraph

  • Modeled agentic AI using LangGraph, where a supervising agent manages reasoning and memory across the interaction lifecycle.
  • The supervising agent coordinates:
    • A RAG agent (Retrieval-Augmented Generation) to handle FAQ and policy lookups.
    • A SQL agent that converts natural language into SQL to check real-time room availability.
  • Leverages LlamaIndex and Sentence Transformers to retrieve FAQ responses based on vector similarity.

5.4 Real-Time Data Pipelines

  • Connects to Neon PostgreSQL for structured data (room types, pricing, availability).
  • Uses Redis for caching and vector lookups.

5.5 Modular Backend

  • Built on FastAPI for scalable API serving.
  • Agent-based orchestration ensures modularity and fault tolerance.

5.6 Guest-Facing Interface

  • Deployed Streamlit frontend integrates real-time chat and booking pipeline.

6. Architecture Overview

End-to-End Flow:
Natural language query → Supervisor Agent (LangGraph) → RAG Agent + SQL Agent → Response Generation → Streamlit UI

7. Results & Impact

  • Interprets diverse guest queries (e.g., “Can I get a king room with a sea view for tomorrow night?”).
  • Vector-based FAQ search yields semantically relevant answers with top-k ranking.
  • Real-time availability checks via SQL queries to Neon PostgreSQL.
  • Seamless user interaction through a modern, responsive Streamlit interface.
  • Deployable via Docker to Hugging Face Spaces and AWS ECS (Fargate).

8. Skills and Tools Used

Category Technologies
AI/NLP OpenAI GPT-4, Sentence Transformers, LlamaIndex
Multi-Agent System LangGraph, Langchain Agents
Backend FastAPI, Redis
Frontend Streamlit
Databases Neon PostgreSQL, Redis
Deployment Docker, Hugging Face Spaces, AWS ECR & ECS
Dev Tools Poetry, Uvicorn, Pytest, Pydantic

9. Future Directions

  • Integrate multimodal search (e.g., images or reviews).
  • Add multilingual support for international guests.
  • Improve CoT prompting for ambiguous guest queries.
  • Enable smart notifications and reservation tracking.
  • Connect to IoT for room service and facility access.

10. Generative AI Capabilities

  • Semantic Query Understanding: Natural language interpretation and slot-filling.
  • Agentic Reasoning: LangGraph-powered agent with memory and decision control.
  • Vector-Based Retrieval: Embedding-driven response generation using LlamaIndex.
  • NL2SQL Reasoning: Translates guest questions to SQL queries for real-time bookings.
  • Multi-Agent Collaboration: Supervisor agent coordinates RAG and SQL agents.
  • Flexible Deployment: Supports deployment to Docker, Hugging Face, and AWS ECS with scalable architecture.

© 2025 Jaber Valinejad. Powered by Docker, LangGraph, and Streamlit. Hosted on Hugging Face and AWS.