Programming Languages: Python, SQL, R, Java, C
Machine Learning & AI: Supervised & Unsupervised ML, LLM’s, Neural Networks, Keras, Tensorflow,
PyTorch, HuggingFace, Productionizing Models, MLflow, Kubeflow
Data Visualization & BI: Power BI, Jupyter Notebook, Pandas, NumPy, Matplotlib, ggplot2, Excel
Cloud & Data Engineering: AWS, GCP, Azure Databricks, Spark, PySpark, Hadoop, ETL, Data Warehousing, Data Pipelines
Business Development & Workflow: Git/Github, Agile, Scrum, Jira, Storytelling, Visualization, CI/CD
Statistics & Modeling: Linear Regression, Bayesian Statistics, Predictive Modeling, Naïve Bayes, K-means,
Decision Trees, A/B Testing
Languages: Portuguese (Bilingual Proficiency), Spanish (Full Professional Proficiency)
Visit my Github.
Bias Beacon
A Google Chrome plugin that automatically detects and highlights bias in news articles. Our capstone team fine-tuned a BERT model with a publicly available bias dataset to detect five different types of bias. Deployed with AWS, Docker, and Replit.
Art Style Analysis
Mapped the features of different art styles. Extracted simple fetures: LBP, HSV, RGB, and FFT. Extracted complex features: VGG16.
Built comparative Random Forrest and Logistic Regressoin classfier models with simple and simple + complex features.
Data Analysis of Harmful Car Crashes
Undergraduate Project where I used logistic regression, area graphs, boxplots, and proportion tables to extensively analize NHTSA (National Highway Traffic Safety Administration) data.
The dataset captures US data, for 1997-2002, from police-reported car crashes in which there is a harmful event (people or property), and from which at least one vehicle was towed. Data are restricted to front-seat occupants.
Headline Evaluation and Style Independency of News Outlets
Measured how accurate news headlines were to their respective article content. Employed NLP techniques such as abstractive summarization, Rogue scores, bleu scores, n-grams, etc. to benchmark results by news outlet. Used similar teechniques to determine style independency of news outlets.
Let's Face It!
Trained and compared multiple models (FNN, CNN, and VGG16) with different parameters to classify emotions of facial expressions, reaching a 60% validation accuracy.