RuiTian

NLP AI Engineer Intern

• Built a multistage document content filter using rerank and lightweight LLM, removing redundant context while preserving 94.2% of effective content and reducing token usage by 30.6%.
• Enhanced zero-shot GRPO with entropy regularization, achieving comparable performance with only 1/4 of the rollouts.

Pharmolix

AI Developer Intern

• Built a Milvus–FastAPI hybrid retrieval system with query rewriting and chunk-level fusion for document QA.
• Developed a PDF-grounded annotation tool enabling accurate chunk-to-document alignment.
• Constructed a LightRAG-based knowledge graph for context-aware retrieval across large corpora.