Well!! How about simplifying your travel planning and giving you a smooth experience when you are still grappling around with the thought about places.
Smooth and Mind Boggling isn't it??
TravelVista presents a unique interface which provides the user an enriching experience where they can write their thoughts or even keywords, and based on this our system will take you through a fulfilling set of choices which you can easily go through and at your fingertip.
System is going to provide the user:
- List of cities
- Hotel or Probable Accommodation Choices
- Nearby site tourist destinations or probable adventure itinerary
Based on these three choices, the system will look out for you and give you a set of itinerary choices for your whole journey and budget as well.
User can have their pick and solidify their travel plans and the best part is you came to this platform with those confusing thoughts and going out with one of the best experiences.
Sounds Fun !!
Do explore our website and get to know TravelVista .... with the power of LLMs.
- External CSV is used for Langchain Model to search upon the necessary dataset, for "cities", "hotels" and "places to visit".
- For all three components, the API is created using Flask, and Langchain is used to search for the query in the external dataset used at the backend.
- BERT is used as an LLM along with Langchain to generate the answers to the user prompts.
- Augmented Dataset is used to give necessary context for the Langchain model to perform better and gain useful insights from the data which we have.
- Dataset is restricted to India right now for all the three contents which we are going to present through our system.
-
City Recommendation: We have manually curated the dataset for city recommendation, using authentic sources like Google Real time Map API and chatgpt, supervised manually by the team. The dataset consists of Destination/City Name, Region, State, User Reviews and a general description (content) of the place. For all the reviews, content are concatenated, we are using FAISS Index to store the embeddings of each entry and then using BERT based Langchain Model for mapping the user query to each embedding. We are using BERT and Langchain since we have to find the best matching places by understanding the users' intent and sentiment and then ranking as well (Matching + Ranking). Therefore, Langchain helps to provide this facility more accurately than just using LLMs like BERT or DistilBERT.
-
Hotel and Places to Visit Ranking: For ranking the hotels and places to visit within the selected city, we have used a custom ranking algorithm. For hotels and places having rating >= 3.7 are considered as good otherwise bad. Good and Bad words are taken out from the reviews accordingly. Each review is matched with the Good and bad word list and a cosine similarity score is calculated and the results are ranked accordingly.
-
Itinerary Generation: Using Map API and LLMs like Gemini, we have carefully curated two suitable itineraries for the user.
-
Engagement Score Calculator: We employ a mechanism where the system gets insights from the user engagements on the webiste, for eg, time he/she spends on a particular destination - taken from cursor hovering time, viewing more images of 1 category, or patterns of dissatisfaction as a feedback for the system. This helps us to understand the user requirements better.
Click the Link to go through this: