Community Bricks

Inspiration

The inspiration for this project came from the challenges faced by individuals with accessibility needs when traveling. Finding lodging that meets specific requirements, such as wheelchair access or braille signage, can be time-consuming and frustrating. We wanted to create a solution that promotes inclusivity and empowers travelers with disabilities to explore the world with confidence.

What it does

The Accessible Lodging Finder is an AI-powered agent that helps users find and verify accessible lodging options tailored to their specific needs. It reviews Airbnb description for accessibility information. The agent provides a ranked list of options using RAG, ensuring users can make informed decisions quickly and easily.

How we built it

We built the Accessible Lodging Finder using:

Data Summary: bright_initiative AirBnb data - specifically, airbnb_properties_information_csv information
Natural Language Processing (NLP): To analyze reviews and identify mentions of accessibility features.
Ranking Algorithm: A scoring system to rank lodging options based on user preferences and verified accessibility features.

Challenges we ran into

Processing Time: Even after filtering down are data to just California sites, it still took hours to process. This made any error in experimental design devastating.
Tokenization: We used a tokenizer that was inappropriate for the model: it lost spaces, split words, and we don't think llama was equipped for this. -Hallucination: We found only 190 descriptions that explicitly mentioned wheelchairs, for example (often in the context of "we aren't accessible"). This probably led to one of our models guessing a lot.

Accomplishments that we're proud of

Developed an NLP model capable of analyzing reviews for accessibility-related insights.
Used vector search, Agents, and Experiments for the first time
Promoted inclusivity by addressing a real-world problem faced by travelers with disabilities.

What's next for community bricks

-Use Reviews: later analysis found reviews significantly more informative: users are a lot more likely to say if their friend or family member was accommodated.

Expand Data Sources: Integrate more lodging platforms and local directories to provide a wider range of options.
Real-Time Updates: Add real-time monitoring for changes in accessibility features or new reviews.
Community Contributions: Allow users to contribute and verify accessibility information to improve the database.
Partnerships: Collaborate with lodging platforms to encourage better accessibility reporting and verification. -Feedback: It would be much more agentic to ask hosts follow-up questions around their space