Leandro J M Machado

I wrote an article that gave explanations on how to efficiently apply machine learning to industry problems.

I designed and led the implementation of a recommendation system for a digital bank that helped its customers set optimal spending limits for each of the 17 transaction categories.

The solution was built in Scala/Spark and leveraged terabytes of customer credit card and checking account transaction data. By using a combination of linear models and user-user collaborative filtering, the solution was robust enough for all classes of customers.

I designed and led the implementation of credit-and-fraud analysis algorithms for an online consumer financing solution.

The solution was implemented in Python and leveraged data from credit bureaus and other alternative data sources to efficiently evaluate customer creditworthiness. The fraud analysis step included facial recognition and matched the provided information with online sources of reputation such as social networks. With this combination of technologies, we could provide a fast-and-light onboarding process for the customer while achieving above market metrics of KS and Gini for default prediction.

I designed and led the implementation of a tool for efficient analysis of A/B test experiments.

It consisted of a Scala/Spark job that took input from several sources—such as impressions, clicks, purchases, and so on—and processed this information to output a number of statistics and hypothesis tests. The methodology included an implementation of the Bag of Little Bootstraps T-test for a robust-and-computationally-efficient analysis. It reduced the time and cost of running experiments by half.

I designed and led the implementation of a real-time recommendation system for cross-store advertisement.

The service was implemented in Scala and made use of variables such as product-content, product-CTR/Conv, and the customer's transaction history to recommend similar products. The solution was built in Scala and consumed data from databases such as Elasticsearch, Cassandra, and Redis. It also served over 30 advertisers and was deployed on one of the top three eCommerces in Brazil.