{"data":{"experience":{"edges":[{"node":{"frontmatter":{"title":"SULI Research Intern","company":"Argonne National Laboratory","professor":null,"location":"Chicago, IL","range":"May - August 2026","url":"https://www.anl.gov/","show":null},"html":"<ul>\n<li>Designed and ran a study of gauge-equivariant neural preconditioners for solving the Dirac equation in lattice QCD; evaluated different model architectures and loss functions with the goal of reducing iterations to convergence on conjugate gradient.</li>\n<li>Delivered a model that cut solver iterations roughly 2x below the state-of-the-art method while reducing its per-system setup cost.</li>\n<li>Proposed three novel architectures and diagnosed why they failed to improve results, tracing the outcome to specific structural properties and narrowing where future gains are likely to be found.</li>\n<li>First-authored the resulting paper with collaborators at MIT and Lawrence Berkeley National Laboratory; presented findings in a poster and a paper (soon to be submitted to a workshop).</li>\n</ul>"}},{"node":{"frontmatter":{"title":"Student Researcher","company":null,"professor":"Dr. Alvin Jin","location":"University of Southern California","range":"August 2025 - May 2026","url":"https://scholar.google.com/citations?user=9Vc6Dn0AAAAJ&hl=en","show":null},"html":"<ul>\n<li>Conducted research with Professor Alvin Jin in the field of topological deep learning.</li>\n<li>Analyzed changes in topological invariants (Betti numbers) across network layers and their effect on inversion ability in predictive coding networks, a type of invertible neural network.</li>\n<li>Developed extensive Python codebase implementing predictive coding networks in JAX and topological analysis using Ripser and NumPy/Scikit-learn.</li>\n<li>Had paper (first-author) accepted to the <a href=\"https://www.tagds.com/events/tag-ds-2026\">2026 Topology, Algebra, and Geometry in Data Science (TAG-DS)</a> conference (also accepted as a lightning talk), the <a href=\"https://www.weightsymmetry.com/\">ICML 2026 Workshop on Weight-Space Symmetries</a>, and presented a poster at the <a href=\"https://zhengchao-wan.com/tda-conference-2025/index.html\">Conference on Topological Data Analysis at the University of Missouri</a>.</li>\n</ul>"}},{"node":{"frontmatter":{"title":"Associate Software Developer Intern","company":"Google","professor":null,"location":"New York, NY","range":"May - August 2025","url":"https://workspace.google.com/lp/solutions/ai/","show":null},"html":"<ul>\n<li>Joined the Workspace AI Platform team, which integrates Gemini into products such as Google Docs, Slides, and Gmail.</li>\n<li>Designed and built an access control system for custom Gems (personalized Gemini instances), streamlining creation and secure sharing across Google Workspace teams.</li>\n<li>Implemented role-based permissions in C++ using Google's internal server framework, managing authorization and routing for millions of potential users.</li>\n<li>Developed Python automation script to manage enum ID generation and updates across a 10,000+ line shared codebase, eliminating merge conflicts and synchronization errors for a multi-engineer team.</li>\n<li>Wrote extensive documentation on my work to ensure correct use and easier navigation in the future; created technical diagrams and professional slideshow to present my work at the end of the summer.</li>\n</ul>"}},{"node":{"frontmatter":{"title":"Software Engineer Intern","company":"Rollick","professor":null,"location":"Chicago, IL","range":"May - August 2024","url":"https://rollick.io/","show":null},"html":"<ul>\n<li>Leveraged Nunjucks to create modular HTML templates and incorporated jQuery to add dynamic functionality and interactivity to web pages.</li>\n<li>Optimized UTM parameters to more effectively identify users and dynamically alter the product accordingly.</li>\n<li>Used KanPlan agile framework built into Jira to more efficiently plan, test, and integrate features, as well as identify and squash bugs.</li>\n<li>Worked in a modern software development environment, creating unique Git branches for each issue and submitting GitHub pull requests for senior developer approval and merging into the final product.</li>\n<li>Utilized Node.js and npm scripts to test and execute code, ensuring quality and functionality before submission for approval.</li>\n</ul>"}}]}}}