Travyl

Inspiration

Out group loves to travel! However, we all agreed that it takes way too long to plan. We take so many variables into the equation when we plan a vacation, and it can result in that process taking several hours or even days. Thus, we wanted to target a simple, natural language to UI itinerary vacation planner. The goal was to take a simple statement and create a planner out of it, with activity, day-to-day, and total budget. We knew we needed a one stop shop for traveling, that was truly intelligent. Most vacation planners and even ChatGPT and many LLM's don't have the capacity to truly take all factors of a trip into account such as safety, weather, distance, availability and more. In the age of AI, we want to come and revolutionize the space with a new way to Travyl!

What it does

Travyl was designed to be simple and intuitive. We want travellers to have the agency to be as descriptive or vague as they would like! For example, "Japan, 10-15th" would reiThe web app would then convert this into a query and key locations/activities, flights, lodging, and restaurants, would be extracted to create an itinerary plan. It was means to be flexible, allowing users to move activities around or delete, creating new activities in their place. It was meant to show budget required to be able to enjoy the week, depending on size of the group, rent lodging arrangements, and help with booking flight tickets..

What it does Travyl was designed with simplicity and intuition at its core. We wanted to give travelers the agency to be as descriptive or as vague as they'd like when planning their perfect trip! For example, a user could simply input "Japan, 10-15th" and our web app would convert this into a structured query, extracting key locations, activities, flights, lodging, and restaurants to create a comprehensive itinerary plan. We built it to be flexible—users can move activities around, delete what doesn't fit, and create new experiences in their place. The platform shows the budget required to enjoy the trip based on group size and helps coordinate lodging arrangements and flight bookings. It's travel planning reimagined!

How we built it

Our main goal was to divide and conquer by splitting the work into four major components: front-end, back-end database, point-of-interest (POI) extraction, and natural language processing. The front-end leveraged AWS Cognito for authentication and was built with Vite and Tailwind for a sleek, responsive experience. The back-end database was powered by Supabase, providing real-time data management. As the project evolved, we integrated additional minor components to round out the functionality. For our infrastructure, we chose AWS as our primary platform since two of our four team members had prior experience with it, allowing us to move faster on the backend architecture.

When a user enters a natural language query, that unstructured data is turned into structured data via a cheap LLM. Afterwards, API's are called using that metadata to return relevant information about the trip and passed into a weighting algorithm we designed. When someone wants to plan a vacation, their decision is a result of an aggregation of all their preferences. Thus, to represent this mathematically we used a graph with weighted edges powered by a cost function. The weights would be rewarded upon favorable conditions (determined by the traveller).

Challenges we ran into

One of our biggest challenges was developing the technique to extract points-of-interest from natural language queries. This component consumed the majority of our 24 hours - it was unfamiliar territory for all of us, but we deliberately chose something challenging because we knew it would push us to grow. It took multiple team members working in parallel to achieve basic functionality, though we weren't able to reach the level of refinement we had originally envisioned. Another significant hurdle was integrating AWS with Supabase for data storage. The team member leading the database work was new to AWS, which created a steep learning curve. However, this challenge became an opportunity - and I came away with valuable experience on a platform they'll use throughout their career!

Accomplishments that we're proud of

We're incredibly proud of what we accomplished within the 24-hour timeframe, and are so excited to take it further. Beyond the technical achievements, we're proud that we did this alongside friends and left as friends - zero conflicts, just collaboration and mutual support for our first hackathon. We all gave it everything we had, and that's something worth celebrating!

What we learned

The biggest lesson? Have realistic expectations. We dreamed big and aimed high! We also learned the importance of scoping projects appropriately for hackathon timeframes and the value of pivoting quickly when initial approaches aren't working.

What's next for Travyl

With the time constraints lifted, we plan to continue refining our POI extraction algorithm until we achieve results we're truly satisfied with. We're committed to deepening our understanding of natural language processing and geospatial data extraction. This is just the beginning - we're going to keep building, keep learning, and keep making Travyl better. The vision is too exciting to let go!