-
-
Conversations on user requirements
-
Project Title
-
All activities displayed on the map and the data frame
-
An activity in the created travel plan
Inspiration
Japan is often referred to as an island nation, where many people are deeply connected to their hometowns and not everyone participates in international activities.
However, post-COVID-19, an increasing number of Japanese are eager to explore the world, especially engineers fascinated by the potential of Data Cloud.
Our inspiration stems from this question: "So, how can we, who might not be seasoned travelers, enjoy our journeys to the fullest?"
What it does
We have developed an innovative application that leverages diverse datasets from the Snowflake Marketplace and the internet, utilizing Large Language Models (LLMs) to craft personalized travel plans tailored to a wide array of preferences and needs.
By using this application, you can enjoy comfortable, safe, and enriching travels.
How we built it
Our idea revolves around using LLMs to understand user preferences and match them with the appropriate datasets.
Initially, Snowflake Arctic collects travel information and requests from users. It verifies if the user's input aligns with the expected queries.
We have pre-arranged a dataset containing details about sights, categories, website URLs, and summaries of attractions.
The sights and URLs are sourced from OpenStreetMap, while the summaries are generated from scraped content, summarized, and vectorized using Snowflake Cortex.
Additionally, we have included crime data sourced from the Snowflake Marketplace.
A travel plan is then crafted based on user interactions and the compiled data.
Snowflake Arctic filters categories and identifies relevant records through completion and vectorization.
Each catchphrase, summary, and image for the selected records are generated using Arctic and Stable Diffusion.
Challenges we ran into
- We encountered limitations in the interpretative capabilities of LLMs; they occasionally failed to accurately grasp user requirements. We addressed this with prompt engineering and refining the user interface, such as employing closed-ended questions from the LLM agent.
- We found that comprehensive, easily manageable datasets containing detailed information about tourist spots were lacking. We mitigated this by using web scraping and LLM-based summarization to supplement our information.
Accomplishments that we're proud of
- We successfully harnessed the power of Snowflake Marketplace and opendata in a novel manner to explore and select sights.
- Our extensive use of Snowflake Arctic, from user interface enhancements to data processing, has been a pivotal achievement.
What we learned
- We gained insights into the specific strengths and limitations of LLM models, notably the high adaptability of Snowflake Arctic.
- Through teamwork and collaborative ideation, we refined our concepts and implementations.
- Creating an application that aligns with our narrative proved to be a highly enjoyable form of self-expression.
What's next for SAKArctic Travel Agency
- We plan to expand our datasets and improve the accuracy of reflecting user attributes, thereby better capturing the essence of the Data Cloud world.
- Moreover, we aim to offer real-time support akin to a concierge service, leveraging data and AI to resolve user issues effortlessly.
Log in or sign up for Devpost to join the conversation.