As a Data Scientist at Deloitte, I have worked with many teams and clients in the cybersecurity area, applying AI to various problems, from anomaly detection in the form of zero-day threats to automatically building out libraries of cyber-specific regulations for individual clients using Generative AI. In addition to this work, I have presented to thousands of technical and non-technical professionals at Deloitte on topics ranging from the Data Science/AI lifecycle to advanced Generative AI use cases in cyber and healthcare.

One of my main responsibilities was researching, developing, deploying, and iterating upon the zero-day threat model created and published by Deloitte [1] [2] [3]. I led the deployment of this model to six different clients across the commercial and government sectors, ranging from transportation and manufacturing to big-tech industries. This involved major refactors of the code to improve throughput while maintaining consistency in the results produced by the pipeline. A major refactor utilized NVIDIA's Morpheus technology and a deep understanding of their RAPIDS and CUDA framework. I became a subject matter expert in the firm and assisted several teams with similar refactors. This framework allowed us to improve the pipeline's throughput from 1 million network flows per minute to 15 million network flows per minute. Using Morpheus allowed our team to switch to a streaming-first paradigm, which further improves the throughput of the larger pipeline (outside of the model itself) by removing the need to store data temporarily. This new architecture and paradigm led to a cost reduction of 80%, as GPUs were utilized far more efficiently.

In addition to maintaining the ZDT model, I led the development of several NLP and Generative AI projects. Each of these projects involved the entire data science lifecycle, from understanding client requirements, researching and performing a literature review, developing minimum viable products, iterating, and deploying full-stack applications that always included a human in the loop to maximize trust and safety in the final product. I utilized technologies such as LangChain and PyTorch to build out an AI pipeline that created a custom library of cybersecurity regulations sourced from Excel spreadsheets and standard regulation documents such as NIST CSF and GDPR. This involved Generative AI technologies such as OpenAI and Anthropic's LLMs to condense and summarize regulations to custom text classifiers to allow teams working with the document to review and correct issues as needed. This reduced the time to delivery by up to 50% according to stakeholders using it in production. Additionally, I am refining a Generative agent using LangGraph that sources multiple types of data including meeting notes and internal documents to create a condensed set of flowcharts that describe processes across a company.

Experience

Technologies

I attended Georgetown University for an 18-month Masters of Science in Data Science & Analytics. A Masters in Data Science significantly increased my technical and managerial skills. I was the head graduate teaching assistant for the Statistics and Probability for Data Science course this fall. My responsibilities include coordinating a group of 13 teaching assistants and three different professors to teach 150+ students. I led the discussion sessions, lab, case study, and homework grading. I was also a graduate teaching assistant for the data structures and algorithms course in the Analytics program. I led the in-class exercise portion of the class and answered any questions students had throughout the class time. Additionally, I was the head graduate teaching assistant for the advanced web application development course in my first semester as a graduate student.

I did research into recommendation systems, especially surrounding image similarity. Most recently, I am working with art collections to provide better recommendations for viewers interested in works of art similar to the one they are viewing. I am writing a paper now that will be submitted to WACV.

Experience

Technologies

During my time as an intern at Deloitte, I developed a custom python package that measures the performance of machine learning and neural network models using a combination of metrics and visualizations to increase the explainability and transparency of models developed at Deloitte's AI Center of Excellence. Additionally, I integrated this package with two production-level projects. I also built out infrastructure around this project, including a GitLab CI/CD pipeline that ran unit and integration tests and automatic documentation generation using Sphinx.

In addition to managing the development process of this package, I also presented the final product to more than 50 people at AI Center of Excellence. I also personally presented to the Deloitte AI Design Council, consisting of AI leaders across the CoE. Finally, I held weekly code reviews with Master Data Scientists at the CoE to get feedback on the product, both from a code quality and architecture perspective.

Experience

Technologies