Hi, I'm Jie 👋
AI Researcher & Algorithm Engineer. Specializing in LLM Agents, Post-training, and Reasoning. Passionate about building intelligent systems.
JW

About

I am currently a Master's student in AI at CUC, with research experience at Microsoft Research Asia and Meituan. My focus lies in LLM Agents, Post-training, and Reasoning. I have published papers at top conferences like KDD and built complex multi-agent systems for game generation. I aim to bridge the gap between cutting-edge AI research and real-world applications.

My Projects

Check out my latest work

Digital Language Museum

Ministry-level language resource platform (Vue/Flask). Developed a BERT/Faiss-powered semantic search engine with re-ranking to optimize high-precision retrieval for large-scale linguistic data.

Vue.js
Flask
BERT
Faiss

Secure Cloud Storage (CUCPan)

Full-stack secure storage with E2EE. Implemented AES-256 encryption and RSA signatures for data integrity. Built with Flask and Vue 3, featuring time-limited downloads and SHA-256 verification.

Flask
Vue.js 3
Cryptography
AES
RSA
Research Publications

My Research Contributions

My research focuses on Large Language Models, Multi-Agent Systems, and Temporal Knowledge Graphs. Here are my recent publications at top-tier conferences and journals.

F

Fine-tuning Multimodal Large Language Models for Product Bundling

KDD Conference

Proposed the Bundle-MLLM framework, utilizing hybrid multimodal tokens and Bundle Prompting for Product Bundling tasks. Achieved 6.4 percent to 11.2 percent SOTA improvement across four real-world datasets.

I

Integrate Temporal Graph Learning into LLM-based Temporal Knowledge Graph Model

Research Paper

Introduced TGL-LLM to bridge the gap in temporal modeling for LLMs by aligning graph embeddings with LLM token embeddings. Achieved up to 15 percent Acc at 4 improvement on the POLECAT dataset.

A

A Comprehensive Evaluation of Large Language Models on Temporal Event Forecasting

Research Paper

Systematically evaluated mainstream LLMs like GPT-3.5-turbo and Llama 2 on temporal event forecasting tasks. Constructed the high-quality multimodal dataset MidEast-TE-mini.

Contact

Get in Touch

Open to new opportunities? I'm always ready to discuss AI research and engineering. Feel free to send me an email directly, and I'll respond as soon as possible.