Inspiration
Travel offers countless opportunities for personal growth, connection, and joy, yet for many people with disabilities, these experiences remain largely out of reach. According to recent studies, nearly 70% of individuals with disabilities do not travel internationally, citing barriers such as lack of accessible infrastructure, uncertainty about support services abroad, and difficulties navigating unfamiliar environments. Even within their own countries, over 40% of disabled travelers limit their trips to locations accessible by car—often relying on themselves or family members for transportation and assistance.
These statistics highlight the profound travel gap faced by people with disabilities—one that is shaped not by lack of desire, but by practical and systemic challenges. Through this project, our inspiration is to break down these barriers, empowering people with disabilities to explore the world with confidence and independence. By identifying key obstacles, developing accessible travel resources, and partnering with industry stakeholders, we aim to make travel more inclusive, enjoyable, and feasible for everyone.
What it does
This project directly addresses a significant barrier faced by travelers with disabilities: the difficulty in finding accurate, up-to-date information about accessible accommodations, transportation options, and attractions. Many people with disabilities choose not to travel, or restrict their travel to familiar destinations, simply because it’s too burdensome to research and verify whether their needs will be met.
Leveraging AI and intelligent agents, this demo streamlines the trip planning process by generating a multi-day itinerary for a selected city. Users can specify their accessibility requirements—such as step-free access, sign language support, accessible public transit, or sensory-friendly attractions. The system then analyzes available data, curates recommendations, and creates a detailed, personalized travel plan that includes accessible routes, activities, hotel options, restaurants, and directions.
How we built it
To bring our vision to life, we combined robust data infrastructure, advanced AI language models, and a flexible agent framework:
Databricks: We use Databricks as our analytics and data processing platform. It allows us to organize and analyze large datasets—including information on hotels, attractions, public transit, and accessibility features—ensuring the itineraries are based on comprehensive and accurate sources. Open Source Language Models (Llama 4): Leveraging Llama 4, an open-source large language model, we interpret user queries, summarize complex travel information, and generate personalized, context-aware recommendations. The model also helps us extract and synthesize relevant accessibility details from various open data and web sources. LangChain Agent Framework: We use LangChain to orchestrate the AI workflow. LangChain enables us to create conversational agents that can reason step-by-step, call appropriate tools or data sources, and adapt the itinerary dynamically according to user needs and feedback.
By integrating these technologies, we built a demo that efficiently gathers accessibility information, generates detailed multi-day city itineraries, and presents directions and suggestions tailored to the unique requirements of travelers with disabilities. This modular and scalable architecture sets the stage for future enhancements, including support for more cities, richer AI conversations, and even real-time travel assistance.
Challenges we ran into
- Creating Actual Bookings: While we successfully assembled detailed, accessible itineraries, implementing real-time booking (for hotels, transportation, and attractions) proved difficult. Most booking providers require comprehensive user data—such as payment information and detailed profiles—which our demo only partially addressed. Navigating multiple booking systems, with their unique APIs and security requirements, added further complexity we were unable to solve at this stage.
- Accounting for a Wide Range of Disabilities: Accessibility is not one-size-fits-all. There are many types of disabilities—mobility, visual, auditory, cognitive, and more—each requiring specific accommodations. Given time and data limitations, we were only able to cover a subset of potential requirements in our prototype. Building a truly inclusive product would necessitate deeper research, expanded datasets, and ongoing user feedback to properly address the diverse needs of the disability community.
- User Interface and Interaction Design: Designing an intuitive, accessible user interface was a significant consideration. People with different disabilities have varying preferences and needs when it comes to technology interaction, such as screen readers, voice commands, high-contrast modes, or simplified layouts. Determining the best modes of interaction (chatbot, forms, voice input, etc.), and ensuring compatibility with assistive technologies, requires specialized UX expertise and user testing—which was beyond the scope of our initial demo.
Accomplishments that we're proud of
- Generating Detailed, Accessible Itineraries: We’re proud that our system can create personalized, multi-day travel plans that account for a variety of accessibility needs. By allowing users to specify their own accommodations—such as mobility requirements or sensory sensitivities—we have demonstrated the ability of AI to move beyond generic recommendations and tailor travel experiences to individual users.
- Extending Beyond Normal Context Windows: We leveraged advanced techniques and infrastructure to empower our language models to process and synthesize information well beyond traditional context window limits. This allowed us to integrate details from multiple data sources and user inputs, assembling comprehensive and coherent itineraries that maintain continuity and relevance over several days of planning.
What we learned
- Using simple for-loops to run our accessibility agent over multiple inputs in a Databricks notebook led to lengthy execution times, highlighting the need for batching or parallel execution.
- Model deployment and serving is straightforward in Databricks, thanks to built-in MLflow integration and Unity Catalog support.
What's next for hack bandits
- API integrations: Connect to Airbnb and Booking.com APIs to search availability, rates, and property details.
- Automated booking flow: Handle user authentication (OAuth), reservation confirmation prompts, and secure payment processing end-to-end.
- Enhanced user feedback: Provide booking status updates, cancellation options, and aggregating multiple platform responses for the best deals.
Log in or sign up for Devpost to join the conversation.