ShuGaze

The ShuGaze Hompeage.
A showcase of the UI, demonstrating its marketplace-like layout.
An example of a shoe a user looked up. The graph shows past prices of that shoe and 5 similar to it, along with the predicted future prices.
Another example of a shoe, this time with the other 5 similar shoes being unchecked, allowing the user to hone in on the shoe they searched.

Inspiration 💡

When's the best time to buy shoes? Should I wait a few days and see if the price goes down before buying? Should I buy them now? What if the price ends up going up after a few days instead? What do I do?

ShuGaze was created as a solution to this dilemma facing online shoppers since the dawn of online markets. Currently, the only data markets allow you to see is the current price of a shoe from one e-commerce site compared to a few others. It lacks the prediction capabilities to determine when one should buy. Thus, ShuGaze helps answer this simple question: will the price of this shoe I want rise or fall within the next few days or weeks?

What It Does 💻

ShuGaze is built as a pseudo-market that allows people to see the different shoes currently on the market. When a user looks up a particular sneaker model on ShuGaze, the current market price of that sneaker is found and displayed in addition to the predicted market price for the next few days. The predictions were made using a machine learning AI algorithm (Random Forest Regressor) trained on shoe price patterns over time.

Our generative AI model identifies the style and contours of the input sneaker model and finds up to 5 other sneaker models that it deems similar in style to the input sneaker model. From this, the user is able to compare both current and future prices of the sneaker model they searched for along with similar sneaker models as well!

How We Built It 🛠️

We trained a machine learning model to predict future fluctuations in the prices of sneakers. This model was trained using past data on sneaker price fluctuations which were taken primarily from StockX.
Built a program that generated similar sneaker models given an input sneaker model. This program was built on an OpenAI-based API. The outputs of the program allows users to find similar shoes as their input shoe.
With both the input shoe and the list of similar shoes generated, the trained machine learning model was able to create outputs for the predicted price of each shoe model.
The front-end is a website that allows users to look up a specific shoe model and see the current and predicted prices of that model as well as similar models. We used SvelteKit for the frontend, and interfaced with our model and OpenAI API calls by writing the backend in Python FastAPI.

Challenges We Ran Into 🚧

Choosing a Model: There were many models to choose from with varying accuracy; we considered doing a time-series analysis, but that approach did not create accurate results, yielding errors of hundreds of dollars. Using a random tree regressor to predict the next price based on the latest price trend reduced error by over 90%.
Data Collection & Cleaning: We primarily took data from StockX, an online marketplace for sneakers. However, we ran into issues with regard to data formatting. Ultimately, we had to clean up and sort our data so that it would be meaningful inputs to the models and programs we built. Though it took a substantial amount of time and effort, we were able to accomplish this successfully.

Accomplishments That We're Proud Of 🏆

Successfully Implementing a Machine Learning Model: We were able to data scrape 500K+ data points and achieved an accuracy of 85.3%.
Coming Up with a Fresh, Relevant, and Powerful Idea: Although it took a good amount of time to come up with our final idea, we are proud of what we ended up with. Our idea has a prominent impact on the world by helping online shoppers save money and has the potential to revolutionize online shopping as a whole.
Completing an Idea Within 48 Hours: The time limit was challenging to deal with considering the scale and depth of our idea, but we still managed to produce a wonderful product despite the difficulty.

What We Learned 📚

Everything Starts With a Good Idea: Our project only took off after we came up with an idea that really spoke to all of us, that being creating a more useful system for buying sneakers.
Data Paves the Path for Solutions: Our idea was a novel one, and we were able to cut through the unknown territory by utilizing the data available to us to the fullest extent, culminating in our creating a meaningful proof-of-concept for our idea.
Teamwork is Essential for Success: For a large-scale project like this one to be successful, we had to know each other's capabilities and build off one another to make our idea a reality.

What's Next for ShuGaze? 🚀

Although we mainly focused on the shoe market for this project, the features and motivation behind it all to find the best time to buy products can be applied to numerous other markets. After all, shoes are not the only product people want to get as cheap as possible. We plan to apply our model we implemented here to many other markets, hopefully creating a wonderful platform that online shoppers can utilize to get the best price on all of their goods. Additionally, we look to find other features that are generalizable to other markets and use those to increase our accuracy with a possibly better model.