About · Research · Experience · Competitions · Tech stack · ORCID
M.S. in Software Engineering @ Xi'an Jiaotong University · AI Algorithm Intern @ Huawei
- 🎓 B.S. in Computer Science, ranked 1 / 156.
- 🏅 National Scholarship · National Encouragement Scholarship.
- 🌟 China College Students' Self-Strengthening Star · “Most Beautiful College Student” of Hunan Province · Outstanding Student Party Member of Hunan Province.
- 💡 I enjoy turning algorithmic ideas into reliable, efficient model systems.
Research themes · Cross-view Geo-localization Uncertainty Learning Multimodal Learning
Exploring · Reinforcement Learning Embodied Navigation Robot Path Planning
| Publication | Role | Status |
|---|---|---|
| IEEE Transactions on Image Processing | First author | Under Review |
| AAAI Conference on Artificial Intelligence | First author | Under Review |
| Acta Automatica Sinica | First author | Under Review |
📍 LUCL studies LLM-guided uncertainty contrastive learning and attention refinement for cross-view geo-localization.
| 🧠 Foundation-model algorithms | ⚙️ AI Infra & inference acceleration |
|---|---|
| Pre-training · Post-training Multimodal / VLM · RL Reasoning · Agents |
LLM inference & serving KV / Prefix Cache · Speculative Decoding Latency · Throughput · Reliability |
- ⚡ LLM inference — speculative decoding, prefix/KV-cache reuse, and prompt compression; reduced first-token latency by approximately 50% in a controlled repeated-context evaluation.
- 🧩 Agent runtime — designed a planner-to-executor workflow for skills, APIs, and subagents with sandboxed execution, structured state, memory, and context compression.
- 📊 Tool-use evaluation — evaluated function calling across explicit, ambiguous, and unsupported intents to improve tool selection and execution reliability.
- 📚 LLM Wiki — a RAG agent for 200+ multi-format documents with hybrid retrieval, reranking, query rewriting, structured tool calls, and safety guardrails.
- 🧪 Large-model training practice — model training and experimentation based on Doubao, covering the standard deep-learning training and evaluation workflow.
- 🛠️ Pimagic Mirror — a Raspberry Pi smart-mirror project selected for a provincial innovation program.
Top-tier algorithmic competition awards.
| Year | Contest | Award |
|---|---|---|
| 2026 | ICPC Shaanxi Provincial Programming Contest | 🥇 Gold Medal |
| 2025 | ICPC Asia Regional Contest · Xi'an | 🥈 Silver Medal |
| 2024 | CCPC National Invitational Contest · Zhengzhou | 🥇 Gold Medal |
| 2023 | ICPC Asia Regional Contest · Shenyang | 🥈 Silver Medal |
| 2023 | CCPC National Contest · Guilin | 🥈 Silver Medal |
More honors · Lanqiao Cup National Final — First Prize × 3 · CCF Algorithm Capability Competition National Final — Silver Award · Baidu Star National Final — Silver Award × 2
Former ACM training-team lead · University / Nowcoder problem setter · Lanqiao Cup problem reviewer
Modeling · Training Fine-tuning RAG Function Calling
Inference · Speculative Decoding KV Cache Prefix Cache Prompt Compression
Algorithms · Data Structures Graph Theory Dynamic Programming Optimization
Build what reasons. Optimize what runs.
Foundation models · efficient inference · multimodal intelligence · embodied AI