Hi, I'm Joy (@le-temps), a graduate researcher at University of Science and Technology of China (USTC).
My core research focuses on Large Language Models (LLMs), Autonomous AI Agent Harness & Memory, and high-performance infrastructure systems.
💼 Industry & Internship Experience
京东集团 (JD.com) · 大模型与搜索算法
2026.03 - 2026.08聚焦大语言模型（LLM）微调、搜索 Query 意图理解、AI Agent 检索增强（RAG）与语义匹配算法研发。
华为云计算 (Huawei Cloud) · 基础架构研发
2025.06 - 2025.09参与华为云原生底层计算基础架构、分布式资源调度与高吞吐算力系统研发，保障云底座高可用与稳定性。
🔬 Research Focus
- Autonomous Agent Harness & Guardrails: Runtime lifecycle interceptors, infinite loop prevention, and tool failure self-healing mechanisms.
- Long-Term Agent Memory: Self-evolving core memory architectures inspired by MemGPT, cross-session memory consolidation, and active reflection.
- Context & KV Cache Optimization: Mathematical analysis of attention entropy, sliding window compaction, and lossless history summarization.
🚀 Open Source Projects
- dsh-plugins-hub: Production-ready plugin ecosystem for DeepSeek Harness (Core Memory, Token Compactor, Code Guard).
- ds2api: High-performance DeepSeek API gateway compatible with standard OpenAI client protocols.
- academic-research-skills: Curated toolkits, best practices, and workflows for academic AI researchers.
📬 Get in Touch
Feel free to connect for academic collaborations, open-source ideas, or engineering discussions: