Dalton Simancek

NLP Graduate Researcher

Responsible Health NLP Research and Evaluation

• Develop privacy enhancements to transformer-based clinical text de-identification model pipelines to reduce the leakage of sensitive patient names by 70% and the impact decrease system false negatives at scale.
• Present research on pseudonymization and error-handling strategies at EACL and the AMIA Annual Symposium.
• Implement a novel NLP fairness evaluation framework utilizing patient demographic information to test de-identification performance gaps across patient subpopulations.

NLP Model Development for Patient Privacy

• Develop a transformer-based de-identification NLP model with the lead researcher, facilitating automated processing for thousands of real Electronic Health Record (EHR) notes for compliance with HIPAA regulations.
• Manage de-identification model assets to standardize configuration files, generate logs, run tests, and streamline installation procedures for batch processing in research and production environments.
• Adopt differential privacy and federated learning frameworks into model training pipelines of clinical text-de-identification workflows.

Cross-Functional Leadership and Mentorship

• Supervise four student submissions to shared tasks for the 9th Social Media Mining for Health Research and Applications Workshops and Shared Tasks (SMM4H) at ACL 2024.
• Lecture as an instructor for the graduate-level Natural Language Processing for Health Data course. Lecture on fundamental ML/AI topics such as Neural Networks and NLP subtasks such as information extraction.
• Collaborate with Michigan Medicine Data Office for further evaluation testing of the de-identification system on unseen clinical text documents.

Data Scientist

Large-Scale Data Analytics Products for Business Understanding and Quality Impact

• Led the development and delivery of HomeKit earnings reports tracking quarterly customer adoption rates, feature usage, and accessory installations for eight company calls with thousands of Apple shareholders.
• Spearheaded development and testing of 30+ always-on dashboards and analytics scripts, delivering real-time and daily analytics updates to dozens of engineers and executives using logging from millions of Apple devices.
• Collaborated with lead analytics engineer to create and validate over 50 new structured logging metrics into 10+ HomeKit software updates for millions of customer devices.

Cross-Functional Collaboration and Mentorship

• Guided HomeKit leadership on the deprecation of further HomeKit support on older models of iPhones and HomePods through analyses of the user population and growth trends, effectively excluding less than 5% of the user base in favor of novel security and framework updates.
• Advised the HomeKit Product team on creating HomeKit user profiles using analytics of device populations concerning HomeKit usage and Smart Home device compositions. Experience guiding HomeKit product direction using novel observations and analyses of the HomeKit user population.
• Mentored data analysts and engineers on HomeKit data pipeline nuances, focusing on opt-in populations, multi-generation software metrics, and internal development build data volatility.

Software Engineering Stability Analyst

Distributed Computing for Software Analytics Strategy

• Independently led and maintained a backend Map-Reduce job for a highly available, always-on software analytics tool used by hundreds of engineers to aggregate MacOS device logs into summary usage statistics.
• Wrote dozens of custom Hadoop/Map-reduce jobs for large-scale batch processing on production grids to track real-time logging and bugs from specific device customer populations.
• Developed Python scripting to automate visualization termination rates across multiple builds from batch processing of customer device logging, reducing the required analyst hours by 50%.

Data Project Leadership and Cross-Functional Collaboration

• Served as engineering project manager for Stability at watchOS Bug Review Boards, collaborating with senior engineers and executives to guide development on concurrent OS updates using analytics-driven strategies.
• Provided mentorship and training to watchOS team managers and engineers on best practices to monitor stability metrics and build their signals into development planning.
• Managed stability project benchmarks by leading cross-functional teams to triage bug fixes and set strategic stability goals for watchOS and its member applications, including Health, Music, Siri, and Fitness Workouts.

Data Analytics for watchOS Software Development

• Independently analyzed and monitored termination rates for over 10 watchOS stability issues, including kernel panics, crashes within the OS, user-facing applications, and background Daemons.
• Curated Top Issue lists for software issues based on aggregate statistics and the impact of bugs.
• Created weekly reports and stability presentation briefings to effectively set and track stability milestones across the software cycle from early internal development through a series of developer builds and into public release.

Continuous Knowledge Sharing

• Led documentation working group on Data Analytics team, setting documentation targets for each sub-team, including DevOps, Web Development, and Data Science, to generate dozens of pages in teamwide documentation.
• Documented dozens of stability metric tutorials and sample calculations for developer audiences and onboarding employees.

Research Area Specialist | Research Assistant

Research Leadership

• Wrote and Awarded a 20,000K grant from Genetic Alliance to film patients and families managing diets with Phenylketonuria (PKU).
• Co-Authored grant and research manuscripts, including two publications: one on a 2x2 factorial design survey relating parent antibiotic use to perceived contagiousness of “Pink-Eye” in Clinical Pediatrics and a qualitative study detailing focus groups with physicians regarding the role of genetic testing in state Newborn Screening programs in the International Journal of Neonatal Screening (IJNS).

Healthcare Data Analysis and Project Management

• Facilitated the mailing and data collection of a national survey of physicians concerning the communication of newborn screening results to parents.
• Administered a parent survey and analyzed the results for a 2x2 factorial design study aimed at relating perceived contagiousness with antibiotic use.
• Co-facilitated physician focus groups concerning the role of genomics in newborn screening and annotated the transcripts following a codebook.
• Translated the patient onboarding curriculum for a Michigan Medicine adolescent weight loss program into a series of instructional videos.

Health Data Visualization

• • Created 30+ infographics to educate physicians, patients, and parents about diagnostic testing, patient experiences, and treatment curriculums related to newborn screening, diabetes, and weight management

University of Michigan Medical School

Coursework and Dissertation Interests

NLP for Health Data | Deep Learning Architectures | Computational Bioinformatics | Distributed Learning | Differential Privacy | Implementation Science Frameworks

University of Michigan School of Information