​CodeBranch's dedicated development team built an AI recommendation system for a B2B marketplace.

​

The recommendation system of this marketplace is used to recommend products to customers based on: the specific needs of the customer and the unique characteristics of that customer.

​

To make the recommendation, the system compares this information with that of other customers who have purchased similar products, who have similar characteristics to the current customer, and who have also provided positive reviews and recommendations of the requested products.

​

This comparison, which is used to make the recommendation, is generated in 32 dimensions to provide the customer with a range of products or services that best meet their needs, thereby creating a better user experience. 

​

The data preprocessing was intense due to the huge amounts of data. 

​

For this project the team added a python microservice running scikit and torch. 

​

Basically data is ingested directly from database to create a plot of points on a 32 dimensions universe. The team used a distance based algorithm (knn) and an “ideal point” strategy to find the closest neighbors to the point, after that candidates were ranked with a normalized distance score and fed to a NN to decide the top 5 products that may be a better fit for the requesting user (represented by the ideal point).

​

The amount of preprocessing for a 32 dimensions universe was intense, since everything needed to be normalized and standardized.