Research intern
2017 - PRESENTUniversity of Buenos Aires- Analyzed the texts of news using natural language processing techniques. In particular, recursive deep models for semantic compositionality over sentiment treebank in order to detect the sentiment of a sentence, dimensional reduction algorithms and topic detection methods were used with the intention of characterizing the mass media bias during presidential elections.
- Focused my study on developing better machine learning techniques which efficiently uses the information of a node and its neighbours. In addition, This new semi supervised methodology will be validated using graphs of biological and social origin.
Technologies: CatBoost, XGBoost, Keras, TensorFlow, Scikit-learn, MATLAB, R, PythonPhD. Software Engineer Intern
2021 - 2021Facebook- I proposed a lookalike model to generate expanded audiences for Facebook and Instagram ads, using user embeddings and graph clustering.
- I implemented a lookalike model to generate expanded audiences in Python.
- I evaluated my proposed machine learning method, and the results showed an improvement in the precision, recall, and conversion score on synthetic and real data.
Technologies: Python, Machine Learning, SQL, C++, Unsupervised Learning, StatisticsPhD. Software Engineer Intern
2020 - 2020Facebook- I experimented with and benchmarked uncertainty estimation methods for ad ranking models of Facebook.
- Implementing Bayesian Deep Learning Models and analyzing more than 20 terabytes of data.
- I improved the normalized entropy score for the ads matching task by 1.6%.
Technologies: Python, Machine Learning, Bayesian Statistics, SQLMachine Learning Team Leader
2018 - 2019Mototech- Developed a forecasting algorithm that predicts the performance of NFL players.
- I managed a group of 4 software engineers and data scientists.
- The results outperform the state-of-the-art techniques by reducing the MSE from 8.541 to 5.576.
Technologies: Keras, TensorFlow, Scikit-learn, R, PythonInvited Professor
2018 - 2019Digital House- Dictated theoretical and applied lectures with code examples.
- I used TensorFlow, Python, and Keras during the lessons.
- The syllabus included the following topics: topic detection, dimensional reduction, embeddings, time series analysis, sentiment analysis, and text analysis/text mining.
Technologies: Keras, Seaborn, Matplotlib, Bokeh, TensorFlow, Scikit-learn, Python, StatisticsData Scientist
2016 - 2017Hexagon Consulting- Implemented different predictive models in order to describe the future financial behavior of bank clients using Python.
- Designed and implemented a recommendation system that uses text reviews in order to recommend a movie using text analysis (topic detection, sentiment analysis, and word embedding).
- Created and develop a software which statistically calculates a personal index of inflation based on a big economic and financial database.
Technologies: Bokeh, Seaborn, D3.js, JavaScript, Predictive Modeling, Scikit-learn, Python