Health Assistant Project: Developed a multi-intent recognition and classification system for online medical scenarios using large pre-trained models and multi-turn rewriting mechanisms. Achieved a 93% accuracy in extracting patient demands, improving adaptability and consultation efficiency.
Science Popularization Assistant RAG: Integrated authoritative medical content, books, and academic papers to optimize recall and relevance, providing accurate health information. Packaged as an agent to support scenarios like virtual consultations, science popularization, and a voice-adapted assistant for elderly users.
Participated in a three-month summer exchange program at the Department of Statistics, Columbian College of Arts and Sciences, George Washington University. During this program, I engaged in academic coursework, collaborated with professors on research projects, and immersed myself in American culture, broadening my international perspective and enhancing my expertise in statistics and cross-cultural communication skills.
Participated in the Kaggle competition focused on generating prompts for large language models, achieving a silver medal in the top 5% of participants.
Competed in the second Chinese Frame Semantic Parsing Evaluation competition on Tianchi, securing top 3 in the open track and top 4 in the closed track.
Awarded the second prize in the 11th National College Student Mathematics Competition, demonstrating strong mathematical skills and problem-solving abilities.
Achieved first prize in the 30th Beijing College Student Mathematics Competition, highlighting excellence in mathematical understanding and application.
Text generation task. Generated training samples for fine-tuning the Mistral 7B model despite limited training data. Utilized a carefully designed few-shot learning strategy to fully exploit the model's potential, achieving a final test accuracy of over 68%.
Extracted frame semantic structures from Chinese text, addressing three tasks: identifying frame information, argument spans, and argument roles. Used pre-trained large models to generate word embedding vectors, coupled with innovative position encoding. Trained a transformer-based model to deeply understand events or situations in sentences, achieving a weighted score of 71.41 across the three tasks.
Employed deep learning techniques to construct CNN and VGG models optimized through transfer learning for efficient recognition and classification of pistachio images. Achieved over 90% recognition accuracy on the test set through meticulous data preprocessing, model training, and optimization.
Python, R, SQL, Stata, SAS
TOFEL 97 marks, GRE 322 marks