Zitao Xing

Shanghai, China

• Conduct research reproduction in AI for Science by reviewing problem formulations, methods, experimental settings, and evaluation protocols.
• Compare reproduced results with reported findings or baselines and document implementation choices, experimental observations, and remaining discrepancies.

Shanghai, China

• Contributed to the development and improvement of Evo Wiki, an AI-native platform for maintainable Wiki and LightRAG workflows.
• Explored knowledge-graph extraction from legal materials and retrieval-augmented generation, with attention to entity-relation modeling, retrieval quality, and traceable evidence.

Shanghai, China

• Supported regional operations by organizing fund-product, market, client-service, and institutional-workflow materials.
• Used Excel and PowerPoint to consolidate, clean, format, and present market information, product materials, and peer-company data.
• Assisted with meeting preparation, client-document archiving, roadshows, and training materials.

Shanghai, China

• Co-developed a touch-first PWA that lets users handwrite people, places, experiences, and works directly on a digital paper interface.
• Contributed to a workflow spanning handwriting transcription, constrained knowledge-graph retrieval, source-backed evidence verification, and traceable annotations.

Shenzhen, China

• Completed intensive training in deep learning and AI for Science, including neural-network architectures, backpropagation, optimization, loss design, model training, and evaluation.
• Built and trained a neural-network model, contributing to data preprocessing, architecture design, hyperparameter tuning, and experimental analysis.
• Studied GPU parallel computing and CUDA fundamentals, including thread organization, memory management, and parallel acceleration for scientific computing.
• Used Python and deep-learning frameworks to analyze how hyperparameters affect model convergence and generalization.

Xiamen, China

• Led a national-level project to build a knowledge-graph-enhanced retrieval system for a Functional Analysis course, coordinating planning, technical integration, validation, and presentation.
• Structured course resources into chapters, concepts, theorems, worked examples, and questions, with explicit relationships among object types.
• Built a course knowledge graph containing 95 objects and 216 relationships.
• Developed a local GraphRAG demo with Neo4j, Qdrant, BAAI/bge-m3, BAAI/bge-reranker-v2-m3, and Flask.
• Validated an end-to-end pipeline spanning semantic retrieval, vector recall, reranking, graph-context expansion, and evidence-chain display.

Xiamen, China

• Investigated numerical solutions of the Poisson equation on sector domains, focusing on corner singularities and nonsmooth structures.
• Helped design a Müntz-type basis method adapted to corner singularities and compared it with algebraic-power bases and a neural-network-assisted baseline.
• Conducted one-dimensional approximation, single- and double-mode Poisson benchmark, and singular-exponent ablation experiments.