Santiago Hernández

Nowadays, working on a data science team involves working with people from different backgrounds, such as mathematicians, economists, actuaries, physicists, and engineers. And being a technical leader of a data science team means that not only you have to ensure that the research, insights, and products add value to the company but also that they are reproducible, maintainable, reliable, scalable, performant, testable, and correct.

For this reason, I started a series of on-site technical pieces of training in computer science and software engineering because the team's data scientist did not have an engineering or computer science background.

The training sessions were on topics like algorithmic complexity, programming exercises, environment setups, data structures, object-oriented philosophy, and software architecture.

On the tech side, this—in conjunction with proper onboardings and code reviews—led to better products and faster development times. On the social side, we were happy to find out that we were unexpectedly making a stronger team by the bonding that arose from studying and debating shared interests.

In the digital marketing industry, performance marketing platforms and their customers have to define the advertisement campaign's goals and a spend model to determine how the customer will be charged (like a fixed CPM, CPC, CPI or CPA).

I worked on a new spend model in which the client and the platform agree on a maximum budget and goals—such as a maximum cost per click or purchase—and the system finds the optimum price to charge the client while achieving the goals and ensuring satisfactory results for both parties.

I built a statistical learning system to estimate the probabilities of mobile app users performing events on applications. The online algorithm processes over 4 million data points per day to continually fit new data.

This was a challenging project as there were different types of apps (like food delivery or flight reservation apps), different types of events (like searches or purchases), and delayed conversion feedback (people usually do not book a flight right after watching an ad).

Before deploying it to production and using it for revenue optimization, it was important to answer a number of questions:
• What is the minimum amount of data required to fit a good estimator?
• What happens if you stop learning from the data and keep using your current estimator?
• How much are you willing to wait for delayed feedback?
• Of all the events in an app which one do you want to predict?

After working on the problem and considering the possible contingencies, the system was rolled out successfully without any issues.

I designed and built the back-end of a mobile game oriented to provide financial education to teenagers.

The game consisted of a simulation of a rapper's career from the ground up, in which the player had to give concerts playing mini-games and take good decisions in order to progress in their career.

The player's manager helps him by giving him advice and suggestions to balance leisure time, taking loans for playing in bigger venues and buying swag to increase the avatar's flow. The project was presented at a contest organized by Itau Bank and Red Hat.