We wanted to create a store in virtual reality in order to present a novel shopping experience to the customers. We looked at different ways to take advantage of the VR shop over conventional offline stores.
What it does
Before the customer enters the store, the customer completes a quick survey. This is used to create a store that is customized to the individual. First, the store collects information about products on Amazon and sorts those products using ML. Then, the store places those products on its shelves based on the clustering and customers' preferences. The customer can shop in this Virtual Reality and then check out the items.
How we built it
There were two main components in the development of our project: Machine Learning and VR.
Machine learning was used to sort the items into clusters based on price, number of reviews, and average rating using k-means clustering. This product information was gathered from Amazon via web crawling. The store places items that will most likely satisfy the customer into shelves that are closer to the customer. We made sure that items from each cluster occupy only one shelf, so that the customer does not run into redundant items in the same cluster repeatedly.
We used THREE.js to build the store in virtual reality. We created shelf models on Blender and imported them into THREE.js. This product is fully functional on any computer since it is web based. It is on the process of being adapted to mobile environment.
Challenges we ran into
Our challenge was building and testing the virtual reality environment. Another challenge was implementing Machine Learning into the store in order to present a novel shopping experience. Finally, collecting data base of the items was challenging, too, due to the inconsistency of information available on each Amazon page.
Accomplishments that we're proud of
We are proud of successfully building a Virtual Reality store and integrating new technology like machine learning. We created a store that has distinctive advantage that cannot be replicated by an offline store: full customization of the store on individual basis. We also made the environment look as realistic as an actual store within the given time.
What we learned
We gained experience on creating a VR environment and ML algorithm. We learned the importance of organization in the development of a complex project.
What's next for Lazy Mart
Even more items can be added to Lazy Mart, not just from Amazon. Natural language processing can be implemented to enhance the accuracy of clustering algorithm and the customer's experience.