"求知若饥，虚心若愚。" —— 史蒂夫·乔布斯

"Stay hungry, stay foolish." — Steve Jobs

• 🔭 Computer Science undergraduate at Chongqing University (CQU), 2023–2027.
• 🔭 Interned with Tencent’s TRS Recommendation Platform as a vector database engineer — worked on GPU-accelerated vector search, online indexing, batch scheduling, and CUDA VMM–based memory management.
• 🔭 Project: Building QQLore, an Agentic Retrieval system for long-term IM data, integrating query planning, hybrid retrieval, reranking, corrective retrieval, and hierarchical memory.
• 🔭 Research Experience: Conducting research on sparse-view 3D Gaussian Splatting — focusing on cross-view geometric consistency, confidence-aware Gaussian optimization, and robust novel-view synthesis.
• 🔭 Silver Medal, International Collegiate Programming Contest (ICPC), 2025 | First Prize, Jiangxi High School Mathematics League, 2021.
• 🔭 Research Interests: LLM reasoning and post-training, AI/Agent infrastructure, and intelligent retrieval, with a focus on efficient training and inference, GPU/KV-cache scheduling, Agent runtime, and retrieval–memory–reasoning coordination.
• 🔭 Seeking master’s, PhD, or research assistant opportunities in LLM reasoning and post-training, AI/Agent infrastructure, and intelligent retrieval, with the long-term goal of pursuing a PhD.
• 🔭 Research Principles: Build strong fundamentals, track the research frontier, develop ideas and insights from concrete problems, stay engaged with the community, and iterate through experiments.

• QQLore — An Agentic Retrieval system for fragmented instant-message histories, featuring query planning, hybrid retrieval, reranking, corrective retrieval, and hierarchical message–event–topic memory for traceable cross-session Q&A and topic tracking.
• drawio-diagram-builder — Coding agent skill for generating draw.io diagrams from natural language.

• CQU-Beamer-LaTeXPPT — Beamer LaTeX presentation template for Chongqing University.
• Exam-Question-Bank — Online quiz platform for CQU party activist trainees.
• Codex-Remote — Bridge between Codex and IM platforms.

• 2026.04 — My Methodology: How I approach learning, research, and building technical understanding from fundamentals.
• 2025.04 — LLM Architecture & Reasoning: Exploring the foundations of large language models, including Transformer architecture, pre-training paradigms, inference optimization, reasoning mechanisms, and scaling laws.
• 2025.06 — LLM Post-Training: Studying alignment and capability enhancement techniques after pre-training, including instruction tuning, RLHF, RLAIF, reasoning-oriented training, and efficient adaptation methods.
• 2025.08 — AI Infra: Understanding the systems behind modern AI, including GPU computing, memory management, distributed training, inference optimization, KV Cache, batching, and large-scale retrieval infrastructure.
• 2025.10 — Agent Systems: Exploring AI agent architectures, including reasoning and acting, tool use, agent orchestration, memory systems, retrieval-augmented generation, and autonomous workflows.