• Lead a student team of approximately four to five members. Alongside core implementation work, I communicate with the client on requirements and technical direction, decompose operator work, assign ownership, review progress and code, and organize testing, reporting, and delivery.
• Lead the design, implementation, and optimization of critical components, resolving compatibility, correctness, and performance issues throughout development.
• Establish and run testing and validation workflows, helping key components move through milestone review and acceptance.
• In biweekly client meetings, I turn vague interface and acceptance language into concrete criteria the team can implement and test. I also clarify boundaries when expanding requirements would put delivery at risk.
• Lead frequent technical discussions, help teammates diagnose implementation issues, and have trained new members on environment setup and the wider team on a consistent Git workflow.

Working setC/C++ · Python · GPU computing · high-performance computing · testing & validation · Linux · Docker · Git

The project is confidential; this page presents only responsibilities and work scope that may be disclosed.

• Independently implemented the experimental pipeline, ran and analyzed the experiments, and prepared the initial manuscript.
• Completed multiple rounds of revision, progressively aligning the empirical results, methodological motivation, and paper narrative.
• The work jointly optimizes ring width, clustering, and node parameters to improve energy sustainability in SWIPT-enabled wireless sensor networks with constrained mobile energy access points.

• This was my first national first-prize award and an important milestone for the long-running three-person modeling team I initiated and organized. We continued competing together in several later contests.
• On a 2,500 × 2,500 grid containing 182,807 weak-coverage points, I owned the enumeration-greedy route for Problem 1 and led Problem 2 end to end, from sector-coverage modeling to direction optimization.
• Expressed the three directional sectors as a polar-coordinate coverage model, used K-means++ to propose primary directions, and implemented conflict repair so every pair of directions satisfied the 45-degree separation constraint.
• Used Julia for the core algorithms, constraint handling, experiments, and visualization, and wrote the corresponding manuscript sections. My work covered the complete chain of modeling, code, runs, figures, and prose and accounted for roughly half of the team’s workload.
• The work received a MathorCup National First Prize. It was also my first time carrying a technical idea through the full cycle of model, implementation, results, and clear written explanation.

• Completed four Python-based machine-learning assignments in a fully English online program without subtitles and received 100/100 on all four assessed tasks.
• While most participants used ready-made library implementations, I chose to build the KNN pipeline—distance computation, neighbor selection, voting, and prediction—from scratch with Python and NumPy.
• Added handling for edge cases such as zero-valued denominators and explained the reasoning behind the selected distance metric.
• Every assignment included a live defense. For the KNN work, the instructor asked me to walk through the implementation line by line in English, including its structure, edge cases, and metric choices.
• The terminology and presentation differed substantially from what I had seen in Chinese coursework. Working through it confirmed that I could learn independently across both a new language environment and a new technical framework.

• Proposed a compact KAN architecture with a single learnable parameter on each edge.
• Currently evaluating it on scientific tabular tasks and Neural ODE settings; the manuscript is in preparation.

• Led testing, deployment, and project coordination for protein-monomer structure-prediction software during the provincial project.
• Later joined the related database system as one of three core contributors, participating in requirements analysis and technical planning while taking independent ownership of the antigen-retrieval module.
• The system was divided into three main components, each owned by one core contributor; my work covered the design, implementation, and integration of my component.
• Co-developed the “Nanobody–Antigen Complex Structure Prediction Software,” registered in China as a computer software copyright on May 21, 2024; listed third among five contributors (registration no. 2024SR0691526).

• Built a working prototype connecting web crawling, filtering, AI-assisted summarization, content organization, and a web interface; tested external access through Cloudflare Tunnel.
• Designed a private clustering approach for content deduplication and organization, inspired by a future-work idea from the MathorCup paper.
• The clustering method is not public; this profile describes the problem it solves without disclosing its algorithm or source code.

• While most course projects used familiar management-system topics, I reimplemented image-filtering principles as a Java desktop application.
• The application could load images and apply multiple filters. It stood out to both the instructor and classmates, some of whom asked to try it themselves.
• The work earned a course grade above 90, the only score above 90 in the class.

• Defined the product direction and designed the color system, page layouts, and complete multi-page interface for a VR equipment storefront.
• One controller graphic was adapted from a free vector asset; the remaining page structure, visual composition, and interaction presentation were my work.
• Interface design portfolio

• Led an early product effort for a WeChat mini-program competition, coordinating the team and moving the idea through to a completed submission.
• The project did not receive an award. It nevertheless taught me the difference between completing a product and making it genuinely competitive.