- Preprocessed 17M+ client tickets with HTML parsing, stopword deletion, link removal, and stemming techniques- Evaluated different embeddings & vectorizations, settling on Word2Vec for semantic analysis of customer messages- Tested K-Means, Agglomerative Clustering, Self-Organizing Map on unlabeled tickets to identify word-sentiment correlations- Implemented Semi-Supervised Learning with ∼1% labels, reducing manual labeling by 99% while utilizing 100M+ data points- Employed Supervised Learning with various Ensemble modeling, achieving 87% F1-Score using Bayesian Optimization

- Connected Generative AI models to the Pega Platform using OpenAI’s open-source API & REST connections to Azure.- Prompt Engineered Chat-GPT to format specific requests and responses to meet varying client needs.- Utilized Azure OpenAI GPT-4 to read resume PDFs and summarize information in accordance with open-ended prompts, allowing for faster sorting through resumes.- Built and presented POC of generalized Azure OpenAI PDF-prompting capabilities for Pega Platform.

Responsible for restarting development of a fraud-detection system meant to identify highly trustworthy non-fraudulent airline ticket transactions.- Cleaned and analyzed 300+ attributes using SparkSQL, identifying 25 critical factors for predicting trustworthy bookings- Developed Support Vector Machine to identify 15-30% “safe” transactions and eliminate false positives- Optimized model using Bayesian Optimization and K-Fold Cross-Validation to guarantee model’s robustness to new data- Presented to leadership a model with error rate of 0-0.0013% allowing $2-4M yearly revenue increase on transaction costs- Trained two interns to take over the project and presented a future path for the next 6 months of development

- Collaborated in an agile team and an 8-team international SAFe- Resolved 10+ critical client-reported bugs and ported to all newer versions of Pega Platform- Implemented and performed unit tests to meet acceptance criteria, using JUnit and Mockito frameworks- Ran performance and endurance tests on Pega Platform using mock API endpoints, simulating >10K customer requests