Nicholas Royal

Possessing analytical and problem-solving skills, coupled with positive attitude. Well-versed in statistical analysis and data visualization, mastering Python and SQL. Capable of transforming data insights into actionable business strategies.

Secret

• Youngstown State University Data Analytics Certification, 06/23
• Technical Excellence Advanced Data Science Program, 04/24
• UiPath RPA Certified RPA Associate (UiRPA) - UiPath, 08/24

• Agentic RAG for Magic: The Gathering: Implemented the Crawl4AI web crawler library in Python for lightspeed web scraping; configured the crawler to look for specific material in the HTML code; prompt engineered gpt-4o to output a summary of the scraped material; chunked the summary into categories and stored it in a vector database; created a front-end for users to interface with using Streamlit; checks Supabase database for entries that best match the user's question and feeds the information into gpt-4o for summarization; provides answers to questions about rules, interactions, formats, and even deck building
• Generative AI for Text Classification: Fine-tuned an NLP model, BERT, to summarize and categorize a given article into one of four different categories; imported the CNN/DailyMail dataset to train the model on; created functions to remove stop words, remove special characters, and normalize the text within the article; used the Tensorflow library to build the machine learning model framework, initialize the tokenizer, and to pad sequences up to the max token length; used the Transformers library to load a pre-trained BERT model and fine-tuned it on articles of varying lengths padding articles that were less than the max token count for the model and feeding it subsections of an article if they were larger than the max token counts; created Docker Files, Docker containers, a FastAPI endpoint, and a virtual environment for efficient redeployment of the fine-tuned model
• Recognizing Handwritten Digits Using AI: Created a machine learning model that accurately recognizes handwritten digits between 0 and 9 with 98% accuracy, conducted data augmentation to aid the model in finding patterns, utilized convolution neural networks and dense layers to train our initial model using Keras, conducted model fine-tuning using Hyperopt and MLFlow
• Predicting Customer Churn: Fit and evaluated multiple machine learning models to predict customer churn including Logistic Regression, Support Vector Machines, K Nearest Neighbors, and Random Forest; evaluated each model using a ROC AUC score to determine which model was the most accurate for our data set; parameter-tuned our Random Forest model using GridSearchCV; found relationships between features and target using graphs and central tendency measures