Research & Work

Academic research, industry experience, and publications.

🔬 Research Projects

AI Agent Smart Campus Assistant Mar. 2026 – Jun. 2026

Core Developer 🔗github.com/csgrace/AI-agent-project

Developed a LangChain-based RAG assistant that builds a complete data pipeline—ingesting, chunking, embedding, and indexing campus documents, with optimized retrieval and citation-backed LLM generation—to deliver reliable answers from unstructured institutional data.
Optimized the RAG pipeline through data chunking, contextual enhancement, and retrieval workflow tuning, cutting average response latency to under 3 seconds while improving retrieval accuracy.
Built a course planning module that ingests curriculum requirements and student credit records to generate personalized semester plans with graduation tracking and credit conflict detection, reducing manual advising effort.

ParroTAO Training Analysis Platform Nov. 2025 – Present

Project Leader (Provincial Key Undergraduate Innovation Program) 🔗parrotao.cn

Led project coordination and product iteration for ParroTAO training analysis platform, driving improvement through internal testing, user feedback collection, and feature optimization. The platform has completed internal testing and is preparing for evaluation with university running teams and coaches.
Designed the front-end architecture and developed key modules—team management, training task coordination, interactive visualization dashboards that fuse training logs with physiological data for performance analytics, and AI-assisted analysis interfaces.
Developed and maintained the integrated project website combining the running community and platform information to support product demonstration and user engagement.

NUS School of Computing — AIoT Summer Program May – Jul. 2025

Front-end Lead Award: First Prize 🔗github.com/csgrace/NUS-AIOT

Led front-end development of an intelligent health management system, designing the system architecture and implementing core functional modules including GPS-based exercise tracking with route visualization, hydration logging, dietary data analysis, sedentary reminders, and personalized settings.
Prioritized user-centric design and interaction experience through iterative usability refinements, ensuring intuitive navigation and accessibility. The system earned First Prize in the program.

General Neural Combinatorial Optimization Model Project Nov. 2025 – Present

Research Group Member

Reproduced Transformer-based neural combinatorial optimization models (Pointer Network, Sym-NCO) within EasyCO, converting raw routing data into unified graph-based representations and implementing reinforcement learning training pipelines for benchmark evaluation.
Evaluated reproduced models on standard benchmark datasets (TSP, CVRP), performing quantitative comparison against reported results to validate implementation correctness, and benchmarked model performance across different problem scales.

University of Cambridge, Department of Engineering Jan. – Feb. 2026

Group Leader, System Design and Engineering Thinking Winter Course

Analyzed stakeholder needs and interactions among students, educators, and AI learning systems, identifying pain points and deriving functional requirements for the BCI solution.
Visualized system interactions and data/information flows via rich pictures and flowcharts, mapping user-system feedback loops and data exchanges to clarify dependencies and refine the learning solution.
Performed reliability and safety analysis using FMEA and Bowtie, identifying failure modes including signal misinterpretation, usability issues, and data privacy risks, and proposing mitigation strategies.

💼 Internship Experience

Meituan Jun. 2026 – Sep. 2026

Agent Training Assistant

Optimized AI Agents via prompt engineering, knowledge base design, and skill tuning, using interaction data analysis to evaluate and improve task quality and decision reliability in complex business settings.
Developed an Agent evolution dashboard to track iteration effects, conducting performance evaluation and data-driven behavioral analysis to enable overall optimization of agent behaviors.
Collaborated with teams to design a four-role RBAC model — Owner, Trainer, Permission Admin, and User — clarifying operational boundaries and enforcing security across the agent platform.

📄 Papers & Patents

LLM-Based Invariant Testing for Software Functional Bugs

ISSRE Accept

Co-author

📄arxiv.org/abs/2607.18711

Proposed LISA, an LLM-based invariant testing framework that leverages execution trace data (API n-gram feedback) to iteratively generate test oracles (API sequences and invariants), improving functional bug detection beyond fuzzing methods.

A Training Plan Generation Method, System, Terminal, and Medium Based on LLM Multi-Agent Collaboration

Under Review — Patent Application No. 2026104381429

Co-inventor

Proposed a training plan generation method via LLM multi-agent collaboration, covering system architecture, terminal, and storage, which ingests user profile and historical performance data to deliver personalized, science-based training plans aligned with individual goals.

🐙 GitHub Projects

CS109 — 2048 Game

Play the classic 2048 game right here!

JavaGamePlayable

▶ Play Now ↗ 🔗 View Repository

CS209A — Computer System Design & Applications A

Try the QQ Farm web demo and explore the Stack Overflow data collection & analysis dashboard.

JavaJavaFXSpring Boot

🌐 View Demo 🔗 View Repository

CS307 — Database
Principles

Full-stack PubMed literature search system: E-R design, 2.8M+ articles, custom bulk import pipeline, Spring Boot + Vue 3 + PostgreSQL.

JavaVue.jsSQL

🌐 View Demo 🔗 View Repository

CS202 — CPU Design

5-stage pipelined RISC processor — hazard detection & forwarding, Harvard architecture, memory-mapped I/O.

RISCPipelineVerilogFPGA

🔗 View Repository

CS207 — Digital Logic

FPGA range hood controller — FSM-driven scheduler, 10+ sub-modules, cross-clock-domain design.

FSMVerilogEGO1FPGA

🔗 View Repository

CS340 — Computing Ethics

Multi-strategy bias mitigation on 646K-record employment data -- adversarial debiasing, MinDiff, 13-metric fairness evaluation.

PythonTensorFlowFairlearnAI Fairness

🔗 View Repository

CS303 — Artificial Intelligence

Othello AI agent combining Alpha-Beta search, MCTS rollouts, and advanced heuristics; Adult Census income prediction pipeline with 9 ML models and 6+ metrics.

PythonMLAI

🔗 View Repository