Case Study: ThredUP

If any job required understanding matching logic, this project was certainly one of them. In our original specification for this job, we were tasked with coming up with an internal matching algorithm to match clothing based on a number of dynamically changing inputs that fed into the database and allowed the program to learn from its mistakes. Setting up servers to handle learning algorithms and simultaneously occurring requests across many different platforms is a challenge for any worthy IT firm. We did rise to the occasion with our original design for the program before these clients decided to switch their business model. In our earliest prototypes of the algorithm we had an index score that was assigned to each possible match. The score was a weighted sum of various relevancy factors which were than analyzed based on user specific data, and past learning outcomes. The score was then compared against the other compiled indexes and suggestions were made the human administrator. Upon feedback from the human administrator, the algorithm quickly adapter to be able make better recommendations and even to start and predict success rate. It is too bad this algorithm was never given the opportunity to go into full scale production since it certainly would have turn quite a few heads in the predictive modeling world.