CV

Profile

MSc Software Engineering candidate at Xiamen University, building context-aware AI systems for long-context management, agent memory, reproducible LLM evaluation, and learning infrastructure.

Education

Xiamen University (985) — MSc candidate, Software Engineering, Xiamen

Sep 2025 – present

Research: context learning, dynamic long-context management, and conceptual-flow analogy.

Henan University (Double First Class) — BEng, Software Engineering, Kaifeng

Sep 2020 – Jun 2024

Selected coursework: Operating Systems (98), Computer Networks (96), Computer Organization (88), Software Quality & Testing (87).

Technical capabilities

  • Languages: Python, C / C++, JavaScript / TypeScript
  • AI & data: PyTorch, Hugging Face, Pandas, Transformers, RNN / LSTM / TextCNN
  • LLM systems: Graph RAG, coding agents, tool use, model evaluation, context engineering, and agent memory
  • Engineering: Git, Linux, Docker, pytest, MySQL, reproducible experiments, and technical documentation

Selected projects

Concept2Fable — Traceable mechanism modeling and structural analogy evaluation

Python · K-12 Knowledge Graph · LLM · 2026

  • Designed the end-to-end research and data pipeline for mechanism-constrained statements, educational narratives, and cross-domain fables, with fine-grained mechanism-to-evidence alignment.
  • Curated 6,541 valid mechanism graphs into 3,183 bilingual three-text units across four subjects: 19,098 texts in total.
  • Designed three structural-analogy tasks and compared 11 models; mechanism-to-text recovery accuracy ranged from 34.1% to 87.1%.

Agent memory and context engineering

Related memory work · Python · Coding Agents · RAG · Evaluation · 2026

  • Investigating adaptive working-memory compression, long-term-memory environments and evaluation, plus multi-source memory conflict detection, resolution, and updating.
  • Building traceable task context from development history, code, and documentation for bug-fix and feature work; the target output is minimal sufficient context with relevant interfaces, similar fixes, test suggestions, and risk constraints.
  • Contributing through source reading, issue reproduction, and testing work around TencentDB-Agent-Memory checkpoint consistency, prompt caching, and cross-platform plugin support.

Multimodal fine-grained sentiment analysis

MABSA · Python · PyTorch · Hugging Face · Optimal Transport · 2025–2026

  • Served as a core research member for literature review, model reproduction, and experimental comparison in multimodal aspect-based sentiment analysis, covering AoM, VLHA, and ADAR.
  • Implemented optimal-transport cross-modal alignment for image-text aspect mismatch, using aspect terms as anchors to connect text tokens with image regions.

ContextLens — AI-powered mixed-reality learning system

Meta Quest 3 · Mixed Reality · Multimodal LLM · ASR · 2025–2026

  • Built a multimodal interaction prototype that brings course video, e-books, voice, notes, and the learner’s physical environment into one learning space.
  • Designed a Context Tree that organizes questions, video frames and captions, book content, and scene context into traceable inputs for grounded LLM responses.
  • Won the Silver Award in Xiamen University’s Challenge Cup entrepreneurship competition.

AI Course DataHub — Public-course learning infrastructure

Site implementation · LLM · Agents · Automation · Static Publishing · In progress

  • Building a public-course pipeline from official resources to transcripts, structured notes, slide explanations, and technical writing.
  • Publishing the resulting learning paths across LLM systems, agents and memory, and AI infrastructure.

High-concurrency cache system

C++ · Concurrency · LRU / LFU / ARC · Undergraduate project

  • Designed and implemented a thread-safe cache framework with a unified interface so replacement policies can be developed, swapped, and extended independently.
  • Implemented the core data structures and eviction logic for LRU, LFU, and ARC, then reduced lock contention through cache sharding and used LRU-k to limit cache pollution from one-off accesses.

TinyRenderer software renderer

C++ · Computer Graphics · Undergraduate project

  • Built a software rendering pipeline from scratch in pure C++, including point, line, and triangle rasterization with basic geometric processing.
  • Implemented TGA image I/O, texture mapping, and sampling, then connected model loading, rasterization, texture sampling, and Phong lighting into an end-to-end renderer.

Open source, community, and honors

  • Maintain reproducible AI learning and research materials through code, tests, READMEs, experiment records, and study resources around Stanford CS336, coding agents, RAG, and multimodal sentiment analysis.
  • Organized an LLM study group around Stanford CS336, agent frameworks, and paper reading, turning discussions into learning paths, code exercises, and technical notes.
  • Built the technical implementation for ClawFight, a multi-agent debate arena at Xiamen’s first hackathon, including agent integration, multi-turn debates, and process-record storage.
  • Participated in Hangzhou 2050 Learning Festival and exchanged AI learning and collaboration practices with open-source communities including Datawhale.
  • Honors: Xiamen University Challenge Cup Silver Award; First Prize, Henan regional College Students Mathematics Competition; Second Prize, Northwest regional College Students IoT Design Competition; Xiamen University Outstanding Communist Youth League Member and Outstanding Party Member.

Links