Paulo Silva

Privately owned colleges in Brazil have a significant problem. Since they are not the top colleges in the country, as the federal universities are, students who join them are usually from lower-income families and tend to drop out frequently.

Students drop out mainly because they face financial hardships, live too far from the campus, can't manage to work and study simultaneously, or even struggle academically and think it's not worth the effort.

Dropping out is a massive problem for the college since the college will miss out on years of revenue from those students. Therefore, it's good for the college to give short-term incentives to hold students in the long run.

With that in mind, I developed a machine learning model to identify the risk and the cause of dropping out. Finally, I provided insight on what incentive the college could offer in trying to hold students.

Some insurance companies demand their clients allow the installation of a tracking device in their car because by tracking the car's location, it is easier to retrieve it. Typically, it takes a while for a theft to be reported, and sometimes it's too late because thieves either remove the tracker or move to a location (usually favelas) where police avoid going to without having a major reason because it's too dangerous for them.

The project I worked on revolved around tracking the users' data, establishing the user's typical behavior using one machine learning model, and then predicting if the car is being stolen using another machine learning model. The objective was to predict these events even before the user reported them to speed up the process of retrieving the car.

For this project, I used Python as the programming language. For the data processing part, we used Apache Spark on the Databricks platform since it was a lot of data, and processing on a single machine was too slow for the requirements (it was time sensitive). The historical data storage was on a MongoDB database, and the API we used to serve the model was Flask.

In Brazil, there's a peculiar industry for selling cooking gas in cans. This industry has operated primarily analogically for a long time. The client would call the vendors closest to them and ask for a delivery, or the vendor would deliver their services by driving around neighborhoods in a truck.

However, once the gas runs out while a person is cooking, they want to have a new can delivered to their home ASAP since not having it may ruin their meals.

With that in mind, the company's business connected vendors and clients through a mobile app. The issue was that these vendors were not used to fierce competition and were very displeased with us.

To calm the situation, we developed a dynamic pricing algorithm using machine learning to maintain the prices at a sustainable level for the vendors while also being advantageous for the clients.

For this project, I used Python for the programming part, Flask to serve my model, and Docker to containerize the model with the API.