I use computing to study molecular behavior and build tools for scientific decisions. My work spans molecular simulation, machine learning, and browser software. I am interested in how models can help us ask better questions—and how evidence can show where their predictions fall short.
Risk-aware decisions from early molecular-dynamics data
Can early simulation data identify runs that are not worth continuing without discarding too many runs that would later meet the study’s endpoint?
PocketStop extracts early ligand- and pocket-motion features, uses supervised machine learning to rank trajectories by risk, and calibrates stop-or-continue recommendations against a user-defined false-stop target. This makes the decision about continuing a simulation an explicit research question, with a cost for stopping a promising trajectory too soon.
My contribution. I defined the problem, ran the simulations and machine-learning analyses, analyzed the data, and wrote the paper as its sole author. I released PocketStop as open-source software.
Current scope. The public package provides a research policy and historical-replay workflow. It does not yet integrate with live OpenMM simulations or include a live pilot. Retrospective recommendations are evidence for evaluating a policy, rather than measured savings from a production deployment.
A guide to the research workflow, not an experimental result. The false-stop target describes the requested risk limit; evaluation measures the observed outcome.
Computational analysis · Ongoing
Connecting Simulation with Experiment
HER2-targeted carrier research
How might attachment to a nanoparticle carrier affect the behavior of a HER2-targeting peptide?
This computational study examines peptide behavior on a polydopamine coating. Molecular simulation offers a way to investigate motion and surface interactions before deciding which questions require experimental evidence. The aim is to understand what the model can reveal about attachment, while keeping the limits of a simulated system visible.
My computational work. Peptide simulation and analysis. The work described here is computational analysis; experimental measurements would provide a separate source of evidence for checking predictions.
Proposed AI extension. I would like to connect simulation features with approved experimental measurements to help recommend which designs or experiments to try next. This is a future direction. A simulation alone does not establish retained recognition, drug efficacy, clinical safety, or successful treatment.
The current work is ongoing. No unpublished laboratory records or preliminary results are included here.
Conceptual research direction
Model peptide–surface behavior
Analyze molecular motion
Compare with experimental evidence
Proposed: inform the next design or experiment
Conceptual schematic. The comparison and AI-assisted recommendation describe a future direction, not a completed or validated feedback system.
Browser software · Published extension
CookieGuard AI
Understanding cookie-related privacy and security risks
How can browser software help people understand potentially risky cookie configurations?
I built and published CookieGuard AI, a Chrome extension that explains potential risks in cookie configurations. Its standalone version combines browser-side classification, rule-based risk assessment, and explanations, giving users a way to examine why a configuration deserves attention.
My contribution. I built and published the extension. The source repository provides the implementation and documentation for inspecting the software.
Privacy choice and limitation. The documented standalone extension analyzes cookies locally rather than sending raw cookie data to a remote analysis service. This description applies to the standalone version; the repository also contains legacy and development components. A flagged configuration indicates a potential risk, not proof of an exploit or a guarantee of complete protection.
My interest in scientific computing began with computational questions related to Valley fever. Conversations with experts helped ground those questions, and computational protein characterization led to my first research paper.
I then learned to build molecular-dynamics simulations with OpenMM and analyze trajectories in Python. Research discussions pushed me to examine what short simulations could reveal, and when continuing a run might add useful evidence.
That question became PocketStop. My current work extends this interest toward the relationship between molecular simulations and experimental evidence. Across these stages, I am interested in making the assumptions behind a computational decision visible, so that a prediction can be examined rather than simply accepted.
Building with others
Community & Leadership
Data Science Hackathon
I organized a data science hackathon for approximately 80 younger students.
Diamond / Conrad Team
I founded a four-person team for the Diamond and Conrad challenges.
I want to connect computer science, chemistry, mathematics, and experimental evidence. I am interested in systems where measurements reveal what computational models missed, while models help decide what to test next.
I would like these tools to report uncertainty, distinguish simulated evidence from experimental evidence, and recognize when there is not enough information to make a useful recommendation. A useful next step would be to evaluate whether recommendations help choose better experiments, alongside their prediction scores. These are goals for future work, and would need validation in the setting where a tool is used.
Outside computing, I enjoy speech and debate, reading, and writing about science and ideas.
Research-learning resources
Molecular Dynamics Foundations
Build an intuition for how atomic motion is computed and what a simulation can—and cannot—tell us.