Seattle, Washington, United States
AI/ML researcher focused on building robust, interpretable systems across security and scientific applications. My research background includes extensive work on AI security and robustness. I've developed backdoor detection and defense methods, explainability-based security metrics, watermarking methods for model ownership protection, and continual learning approaches with robustness guarantees. Currently building AI systems that help researchers optimize experimental design through structured data extraction, LLM-based protocol analysis, and predictive modeling. My work focuses on creating reliable, interpretable AI tools that augment human expertise in data-driven research environments.
• Building core AI systems behind Pando, an AI platform that helps plant scientists design and optimize tissue culture experiments by learning from prior data and guiding experimental decisions • Working closely with plant scientists to translate experimental knowledge into user-facing tools • Expanding an LLM-based protocol extraction system to generate structured datasets of experimental conditions and outcomes from scientific literature • Developing predictive models that estimate protocol success and recommend experimental conditions likely to succeed for a given goal • Designing and implementing a DOE-based recommendation system with a custom search algorithm over user-defined constraints to identify feasible designs and guide minimal corrections when inputs are incompatible
• Designed a structured protocol data model (built on the company’s “Plant Culture Language” ontology) to represent plant tissue culture experiments in a standardized, ML-ready format • Built an LLM-powered pipeline to extract experimental details from scientific papers into structured data, enabling downstream modeling and analysis • The above pipeline enabled product features where users upload papers to extract media conditions, and internal pipelines that generated a large-scale dataset of experimental media from literature • Applied Design of Experiments (DOE) methods and Bayesian optimization with lab scientists to design and refine somatic embryogenesis experiments, improving efficiency and reducing trial-and-error • Contributed to early development of an ML-based recommendation system for experimental design
• Leading NSF-funded research focused on enhancing the security and robustness of Graph Neural Networks (GNNs) through watermarking techniques and backdoor detection methods. • Developed a groundbreaking watermarking framework for GNNs, enabling model ownership verification while preserving model performance and resistance to adversarial manipulations. • Designed and implemented a novel detection framework that integrates GNN explainability mechanisms with multiple custom metrics, significantly improving the detection of backdoor attacks. • Ran extensive experiments on benchmark datasets to validate both watermarking and backdoor detection approaches, demonstrating their effectiveness against a range of adversarial threats. The results reinforced the robustness of these methods in safeguarding GNNs. • Spearheaded research initiatives that addressed critical gaps in AI security, advancing ownership verification and adversarial defenses for GNNs, while aligning these efforts with broader advancements in AI robustness and security.
• Delivered comprehensive instruction for 'Introduction to Computer Programming (CS 105)' and 'Introduction to Computer Programming for Engineers (CS 104)', guiding students through foundational programming concepts. • Taught essential elements of computer science and programming, focusing on C/C++ and MATLAB languages, encompassing key areas such as syntax, iteration, and function development. Conducted both in-person and virtual recitation sessions, managing approximately 30 students per section, providing tailored support and feedback. • Responsible for assessing student assignments, ensuring fair and constructive grading, and hosting office hours to offer additional academic support and clarification.
• Developed prototypes and custom parts for entrepreneurs and engineers, employing manufacturing techniques such as CNC & 3D printing. • Conducted tensile and compression tests to determine material limits and identify potential manufacturing constraints. • Built and installed art pieces at the WNDR Museum, Lululemon, Paylocity, Elkay, and the 900 North Michigan Shops.
• Made and served drinks to audience members in the Second City's three main theaters • Served hundreds of customers per night