Sean Steinle

Evaluated LLM and AI system outputs for accuracy, reasoning quality, and reliability. Built and maintained automated evaluation/testing pipelines to measure model quality across releases. Reviewed model-generated code and assessed prompts/responses by documenting failure modes and recommending improvements. Checked training and evaluation data quality for consistency and labeling errors.

Improved generalization of reinforcement learning agents using symbolic goals and curriculum learning. Guided development, testing, and deployment of a multi-language package (Python, Julia, PDDL) using GitLab CI/CD. Built an ETL pipeline to transform 6 million raw patents into actionable intelligence. Advised a team of college students developing a cloud-based security tool.

Served as principal investigator on a project to generate simple, trustworthy scores for large language models. Designed a multimodal foundation model for cybersecurity applications and led evaluation of federated learning models for classifying malicious network activity, focusing on scalability and robustness. Managed a code repository for a team of developers and led multiple migrations to new remote code services. Supported the department intern program and helped hire interns.