TraverGo
Discover Your Perfect Stay with Our Travel Advisor: Tailored Hotel Matches, Effortless Conversations with Our Chatbot, and Real-Time Insights for Your Ultimate Hotel Search Experience.
Created on 11th February 2024
•
TraverGo
Discover Your Perfect Stay with Our Travel Advisor: Tailored Hotel Matches, Effortless Conversations with Our Chatbot, and Real-Time Insights for Your Ultimate Hotel Search Experience.
The problem TraverGo solves
Finding the perfect accommodation for your long-awaited vacation can be a difficult and risky challenge. Some of the primary problems with the current system are,
- Inability to address travelers' personalized preferences: Current websites and apps don’t give the ability to the traveler to easily search through a multitude of hotels, descriptions, and reviews to find one that fits their needs.
- Time it takes to research hotels: Traverls need to spend lots of time researching the specifics of a hotel which is time-consuming and in the long run has the traveler forget about the important bits.
- Conversational style interaction: Speaking to a human assistant is time-consuming because of hold timings as well as it being an inconvenience which leads to travelers being discouraged to do the same.
- Revenue loss: Because of the above reasons, hotel companies are unable to personalize their hotel or rooms for their customers ahead of time. Addressing the needs of the customers and their concerns helps in matching customers to hotels.
We solve these problems by implementing and giving the following features,
- Sentiment Analysis: Analyzes reviews and generates comprehensive ratings based on sentiment analysis, allowing travelers to make informed decisions.
- Vector Database/Neural Search: We give the ability to match user queries to different hotels based on the hotel description and reviews.
- Traversaal AI API Integration: With the Traversaal AI API, real-time information is fetched about nearby locations, allowing recommendations for nearby attractions, restaurants, and transportation options.
- Decoder Model: A decoder is used to understand why a particular hotel is the best match for the traveler's needs, providing insight tailored for each traveler.
- QnA Chatbot: Added a feature for users to ask for further questions and clarifications on details of the hotel. Also integrated with Ares API to address real-time information.
Challenges we ran into
The biggest challenge was to learn all of the different frameworks and tools that we needed to implement the desired features. In particular, learning data manipulation with Qdrant and integrating the backend features with Streamlit was particularly time-consuming. Qdrant was used to create our personalized vector dataset, data processing, and addressing the hotel reviews through sentimental analysis. Much of our time was spent cleaning the data into leaving with only the ones that would benefit with the model’s recommendation. In addition, some parts of the data were either missing or casted with the wrong type, leading to issues in implementing what was originally considered straightforward functions. With such a small detail being so easy to glance over, we spent much of our time debugging edge cases which took away time to develop the important features of our model.
Streamlit allowed for us to create session states, which allows for variables to be shared throughout (re)runs and user sessions. In our case we used a decoder model to generate explanations for each of the hotels we chose. To make the generated text persistent, we had to create multiple session states for each of our hotels and save the explanations generated in them which were later retrieved in app reruns.
While developing, we also noticed the very prevalent slow run times while trying to check the performance of our model. OpenAI API and Ares API took time to generate due to their large traffic. Being on a limited token usage for both of our API proved to be fairly nerve-wracking as well, especially knowing that we can only test our model a limited number of times. Regardless, our team did our best to persevere through the difficulties.
Tracks Applied (1)
Traversaal AI Track
traversaal.ai