Mauricio Santana-Vargas
Director Data Science @Deserve
Signup · Get unlimited contacts
WORK HISTORY
Director Data Science @Deserve
Palo Alto, CA, US
Managing teams of Data Scientists in the US and in India that oversee credit models, financial reporting, compliance reporting and auditing as well as marketing.Coordinating across different teams, including Finance, Accounting, Compliance, Engineering, Marketing, Operations to understand and prioritize data science models, data dashboards, reports, information exchanges and ad hoc data requests.Coordinating the automation of recurrent data modeling and requests through the team\'s self- designed libraries in Python, R and AWS.
EDUCATION
The London School of Economics and Political Science (LSE)
MSc, Economics
Tecnológico de Monterrey
BSc, Economics
Carnegie Mellon University - Tepper School of Business
MSc, Finance
ABOUT MAURICIO SANTANA-VARGAS
I am a professionally trained Economist through several Master\'s degrees mainly working in banking and finance. I have managed teams across the world doing tough machine learning problems, mainly classifications that drive the business profitability. The models of my teams have kept in check losses (1% of balances) of multibillion dollar credit product programs. The models addressed both, credit and fraud. We have also done causal inference, LTV (Lifetime value) estimations and multipoint attribution for marketing. Earlier in my career I had done catastrophic modeling extreme losses estimation.A key success point has been communication. In order for our models to deliver that kind of performance I have had to work with Marketing, Accounting, Operations, Credit, clients, providers and CEOs, all of them asking me questions and talking from their point of view. I have to earn everyone\'s trust and communicate clearly. And across countries to boot. That was not easy. A second success key has been the amount and quality of data needed for the models. My teams and I have specialized in distributed storage and processing including AWS redshift databases (that I designed) as well as datalakes (including databricks) and PySpark environments for processing. Then we use XGBoost and Neural Networks for classification. After all of that, the rest of my anxiety (80%+) comes from assuring excellent data quality.But I like to stay current and I do program in Python, C, C++ and other languages as required. On top of models, I have coded myself data pipelines, automated job scheduling through airflow and Jeeves as well python API calls and CI/CD (continuous Integration continuous development) of models. How close to the code? I have had to update Unit Tests inside docker libraries via bash in python using mock for some of the data pipelines my teams used as an example.I also do have other endeavors in Computer Graphics, entrepreneurship and strong volunteering. With the latest AI developments I have worked in helping teams with minimum value proposition designs that focus on vision classification and trying to reduce hallucinations. Mainly through GPU training and LoRA (Low Ranking Adaptation). I have worked with 3d data, including point clouds, voxels, signed distance, deep images, hdr images and 3d meshes.I get bored quickly and must do something (usually mathematically complex) or help others to pass the time!More than happy when anyone reaches out and says hello! 🤗
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.