Multimodal LLM Post-Training · Reinforcement Learning · Retrieval
Software Engineering student focused on multimodal model post-training and reliable evaluation.
Open to research and algorithm internship opportunities.
- AutoConcept — First-author, training-free concept-guided reranking; accepted at PRICAI 2026 (CCF-C). [arXiv] [Paper]
- Multimodal UI Modeling — Post-training and evaluation for UI understanding and interaction models.
- AMEND — Multi-agent, constraint-aware text detoxification through specialized critics and minimal repair. [Report]
- Ant Group — Multimodal LLM post-training intern; built
verl-based training workflows and evaluated UI grounding models. - R&L Lab, Nanjing University — Integrated 127 algorithms and built a processing pipeline for 70,000 road-scene images.
- Incoming M.Eng. student at Southeast University; B.Eng. candidate at Soochow University, top 6% by GPA.
- Python · C/C++ · PyTorch · verl · SFT · GRPO · VLM · GUI grounding · distributed training
- NOIP National First Prize · CSP 370 · CET-6 659