Peixin Li

Developed an LLM-based sequential recommendation framework that integrates collaborative filtering with semantic language modeling to address cold-start and sparse-interaction challenges. Combined ALS, RQ-VAE, and Qwen2.5 with parameter-efficient fine-tuning and constrained decoding, achieving 8.8% higher HR@10 and 6.9% higher NDCG@10 than SASRec. The project was completed as an undergraduate thesis and resulted in a registered software copyright.

Developed a multimodal AI system for rice disease diagnosis and intelligent consultation based on Qwen3-VL-4B, integrating QLoRA fine-tuning with LangChain + FAISS RAG for image-based diagnosis, expert knowledge retrieval, and contextual question answering. Optimized deployment through 4-bit quantization and adaptive inference, covering the complete workflow from data preparation and model fine-tuning to Web-based system integration.

Collaborated in interdisciplinary work with e‑commerce majors, leading industry research, user research, visual design, and development for an AI‑powered health‑management solution, earning Second Prize, South China Division, China Collegiate Computing Contest. Addressed insufficient personalization in chronic‑disease management via market research, product planning, business‑model design, and the development of an AIGC agent‑based solution for patients with diabetes and high‑risk populations.