Kevin Grazel

Senior Machine Learning Engineer – CVSHealth

July 2024 – July 2025

• Led the design and deployment of robust, production-grade ML systems to detect and block botnet web traffic in real time, training on 2TB of data and building scalable, maintainable ML pipelines.
• Developed and optimized a denoising variational autoencoder for anomaly detection with advanced deep learning architectures and rigorous unit and integration testing.
• Engineered custom tokenization and transformer-based embedding pipelines to convert sparse web traffic data into dense, model-ready tensors, reflecting DSP-style data processing.
• Championed MLOps best practices by automating model deployment, monitoring, and retraining workflows for continuous production quality.
• Collaborated cross-functionally with engineering and product teams to integrate ML features into production systems, supporting ongoing feature development and bug triage.

AI Engineer – Parker-Hannifin

February 2024 – July 2024

• Led end-to-end development of a retrieval-augmented (RAG) chatbot platform, architecting Python-based backend systems for scalable, production-ready deployment on Azure.
• Integrated ML models and MongoDB vector database systems into enterprise environments.
• Instituted CI/CD automation and code quality standards, mentoring junior engineers and fostering a culture of maintainable, high-quality software development.

Senior Data Scientist – PlanetX

November 2022 – July 2023

• Tripled the number of products available on PlanetX website using semantic NLP-based and fuzzy text-matching methods to improve record linking logic.
• Built scraper bots with Selenium and Python to source alternative data and link unmatched brands to companies.
• Expanded machine learning infrastructure on AWS EC2 and S3 services.
• Wrote ad hoc SQL queries to investigate core model statistics and fine tune scoring distributions.
• Worked closely with Product and Engineering teams to identify and fix anomalies and address user UX feedback.

Data Scientist – MUFG Investor Services

November 2019 – November 2022

• Investigated feasibility of forecasting investor capital movements.
• Used dbt to build data warehouse backend ETL processes in AWS.
• Wrote SQL queries to obtain raw data; cleaned and structured data to be used to train machine learning models.
• Built model training suite to rapidly train and track performance of hundreds of models.
• Used Bayesian optimization to tune LightGBM model hyperparameters.
• Achieved a lift value greater than 5.0 in targeting accounts likely to redeem their investment.
• Developed ESG reporting product to expand core fund administration business into rapidly growing ESG sector.
• Designed and led development of backend data architecture as well as frontend PowerBI user interface.
• Planned product roadmap toward further business cases given MUFG's unique position as a fund administrator.
• Data backend of ESG product used as basis for new company-wide data architecture.
• Expedited Ops Team account deduplication project by 10x using fuzzy matching to identify duplicate accounts.
• Developed a proof-of-concept system to extract information from invoices to reduce invoice processing workload.
• Parsed PDF files to obtain raw text; trained SpaCy NER (Named Entity Recognition) models to process text.
• Established performance baseline and foundation of project for further development.

Senior Risk Consultant – Process Risk, LLC

April 2016 – November 2019

• Facilitated Process Hazard Analysis studies and Management of Change discussions.
• Led teams of three to ten engineers and operators, serving as project consultant and client point of contact for projects at industrial chemical process complexes across the U.S.
• Cut project lifecycle time by 50% by writing Visual Basic macros to automate processes and deliverable creation.

Data Science Fellow – Metis

July 2018 – December 2018

Metis is a highly selective, immersive, full-time bootcamp in which I developed several data science projects, including:

• Time series analysis of federal economic data to forecast probability of economic recession
• Deseasonalized regression model to forecast subway ridership load by weather conditions
• Designed voice recognition algorithm to identify speaker in audio clips of presidential speeches
• Investigated trends in the focus of Congressional legislature using Latent Dirichlet Allocation (LDA) topic modeling.