We were inspired by our experiences in buying and selling products on e-commerce platforms, where we consistently came across sellers driving away potential buyers with poorly written or misleading descriptions. We wanted to fix that so that both buyers and sellers can provide and obtain goods and services easily.
What it does
Have you ever wanted to sell an item on eBay, but felt like you didn’t have the time to make the listing? Now, listing is as simple as just taking a picture and letting eBAI do the rest! eBAI uses state-of-the-art computer vision technology to identify your item exactly. It then generates all of your listing details for you (item description, condition, price, picture, etc.) and posts your listing directly to your eBay account. As we like to say, “We'll post it to your eBay account - all you need is a picture!”
How we built it
We developed the eBAI (eBay AI) machine learning components using Google Cloud Platform AutoML Vision, Google Cloud Storage, Microsoft Azure, and OpenCV.
We started by creating our own robust computer vision dataset using Microsoft Azure Cognitive Services and OpenCV, generating over 1,500 images with 5 labels. This dataset, when run through our neural network, achieved an impressive average precision of 0.963.
Simultaneously, we set up **AutoML Vision within the Google Cloud Platform. We used AutoML Vision to train and deploy our model.
Also, simultaneously, we used the eBay Offer API to create Swift functions that would allow us to post to eBay through our iOS application.
We then created the front-end of our application using Swift, and connected it to our back-end machine learning and eBay components.
Challenges we ran into
We ran into plenty of challenges along the way, primarily with Google Cloud Platform Authentication and incorporating our API calls to AutoML from our iOS app.
Accomplishments that we're proud of
We are proud of the accuracy of our neural network and the meaningful tags generated by our model. Using them we were even able to determine the condition of the product a seller is posting on eBay.
What we learned
We learned how to scrape 1500 images using Azure's Cognitive Services and how to implement it with eBay's API for getting similar inventory and creating tags for posting images of the uploaded products. Also even though it was arduous task, we learned how to authenticate Google Cloud Platform for our custom API calls.
What's next for eBAI
Our code is production ready - the only hurdle between our current product and the marketplace is the time it takes to collect and train additional data. Given the necessary GPU resources, that could be completed in as quickly as a few weeks or months.