Inspiration
EasyNav was inspired by the desire to support those who face additional barriers specifically neurodivergent individuals who already navigate daily challenges. Travel can be overwhelming from building the perfect itinerary to finding accessible accommodations, the process is often stressful and exclusionary. EasyNav aims to change that by offering simplicity, comfort, and confidence transforming travel into an enjoyable and empowering experience.
What it does
EasyNav is an AI agent that analyzes Airbnb listings to create a personalized shortlist for users. It focuses on listings that meet these specific criteria:
- Quiet accommodations to minimize sensory overload.
- Self check-in for independent, stress-free arrival.
- Low interaction environments by prioritizing less dense, quieter neighborhoods
How we built it
We built EasyNav using Databricks functionalities, including vector based indexing and Databricks Maverick to create a responsive, personalized AI agent.
Quickstart Template: We began with the Databricks quick start template and tailored it to meet the specific needs of neurodivergent travelers.
Data Preparation: We filtered, cleaned, and analyzed the Airbnb dataset for relevant information, focusing on keywords such as: quiet, not noisy, calm, self check-in, and no interaction environments.
Vector Search Setup: Carefully defined the columns that would be relevant for vector search, including descriptions, details, number of reviews, ratings, and reviews.
Conversational Interface: Finally, we developed a chatbot and a user-friendly interface to make it easy for travelers to interact with the data and discover the best listings for their needs.
Challenges we ran into
Some challenges we ran into were time, inexperience with LLMs and AI agents. We spent a good portion of our time reviewing documentation and learning the basics of LLMs.
Accomplishments that we're proud of
We’re proud of building a functional proof of concept especially since we were new to these tools. Applying vector search to the dataset was a key milestone, and seeing it actually work (even if just slightly!) was a big win for us.
What we learned
We learned more about LangChain, the concept of vector search, and how it enables more dynamic, relevant results. Instead of relying on keyword-based training, we learned to leverage vector representations for more nuanced understanding and retrieval.
What's next for EasyNav
The next steps for EasyNav include expanding our search functionality beyond San Francisco, refining the vector search to capture even more nuanced user needs, and adding features to the chatbot interface to make interactions smoother and the overall experience even more intuitive.
Log in or sign up for Devpost to join the conversation.