Accessible Travel Agent

Databricks Data + AI Summit 2025

---

Overview

The Accessible Travel Agent demo empowers travelers with disabilities to plan multi-day, fully accessible trips in seconds. Leveraging Databricks for data processing, Llama 4 for language understanding, and LangChain for agent orchestration, this project generates personalized itineraries based on user-specified accessibility needs.

Features

• Custom Itinerary Generation: Specify accessibility requirements (e.g., step-free routes, sign language support, sensory-friendly attractions) to receive a detailed travel plan.
• Data-Driven Recommendations: Aggregates hotel, attraction, and transit data from Delta tables in Unity Catalog.
• Scalable Model Deployment: Built-in MLflow integration for seamless model logging, versioning, and serving in Databricks.
• Modular Agent Framework: LangChain-based architecture allowing easy extension with new data sources or booking APIs.

Architecture

1. Data Layer: Delta Lake tables stored in Unity Catalog for hotels, attractions, transit, and accessibility features.
2. Notebook Workflows: PySpark notebooks preprocess and join data, exposing it to the agent.
3. Agent Layer: LangChain agents call custom tools to query Databricks SQL and format responses.
4. Model Layer: Llama 4 model handles natural language understanding and itinerary synthesis.
5. Serving Layer: MLflow model registry and serving endpoints provide real-time itinerary generation.

Prerequisites

• Databricks workspace with Unity Catalog enabled
• Cluster with access to Delta tables and MLflow
• Python dependencies: langchain, databricks-sdk, mlflow, pandas
• API credentials (for future booking integrations)

Getting Started

1. Clone the repository

   git clone https://github.com/g20yr/dais_hackathon_2025.git