For us, when we go to new places for travel, we never really know what to do. Obviously, we can search "things to do in X," but that just gives an overwhelming list of locations that we may or may not want to go to. We wanted an application that could provide us a streamlined list of locations that we know we would want to go to. This gave rise to our creation: travelMe.
travelMe is a travel guide application that generates travel plans specifically catered towards the user. It uses generative AI to learn the user's travel preferences through their travel history, as well as self-inputted and real-time location data. For example, if the user has historically been to a lot of coffee shops, travelMe's travel guide will curate a travel plan filled with cafes, bakeries, and other similar destinations. Taking user-inputted and real-time data into account also means travelMe's travel plan provides the most optimal route and locations for the user, locations that would avoid obstacles such as weather issues and would be closer to the user's staying location. travelMe solves the problem of having to decide one by one which locations to go to by providing a travel catered towards to user.
To create travelMe, we used React, Javascript and HTML/CSS for the front-end, Flask, and Python for the back-end. The core of our project is a machine learning algorithm that combines generative large learning models with the tools of finetuning and Retrieval Augmented Generation to make it powerful that any normal model like chatGPT. We trained the model upon hundreds of well-defined trip plans, as well as scraped multiple well-known and established websites such as TripAdvisor for accurate descriptions and reviews. All of this information is stored in our database where we use cosine similarity search to extract the most relevant information to enhance our response and trip planning for the user. We also incorporate real time data as context, such as the project weather of the location the user is visiting as well as any new events happening around the area through multiple API's like Yelp and Eventbrite. We also use a self reflecting AI Agent to constantly keep track of both the user's long term and short term preferences to enhance personalization. All of this is used in tandem to craft the perfect trip.
The main challenge we ran into was connecting the front-end to the back-end, and specifically parsing JSON formatted data into strings and back.
We are proud that we were able to build an application that solves a specific but common problem that many of us have ran into, incorporating our knowledge of machine learning models, frameworks, and back-end and front-end development strategies.
We learned about the details in connecting front-end and back-end, and states.
The next step for travelMe is to incorporate a map API so users can view a visualization of the travel plan they are going to be taking.
Log in or sign up for Devpost to join the conversation.