Highly accomplished AI/NLP Algorithm Engineer with a Master's focus in Natural Language Processing, specializing in large language model optimization, algorithm development, and data-driven solutions. Proven ability to enhance advertising creative generation, improve search relevance, and drive significant business impact through advanced machine learning techniques and robust research methodologies, as evidenced by +14.44% offline quality and +0.56% CTR gains.
Advertising Creative Algorithm Intern
Shanghai, Shanghai, China
→
Summary
Led the development and optimization of AIGC-driven advertising creative algorithms for Xiaohongshu, significantly enhancing material generation efficiency and online advertising performance.
Highlights
Developed and deployed multiple AIGC models for advertising creative generation, achieving an average +14.44% improvement in offline quality metrics, a 2.53x QPS increase, and online gains of +0.56% CTR, +0.27% CPM1, and +1.10% cost efficiency.
Designed and implemented a multi-source LLM distillation strategy, generating over 800,000 high-quality training samples through multi-stage filtering and online data feedback.
Optimized model training with SFT cold-start and DPO alignment, boosting model output diversity by +2.78% and rejection recall to 96.18% for online service safety.
Spearheaded research into black-box reward modeling and RL optimization for LLMs, achieving a new Pareto front with +2.36% average quality improvement in offline evaluations.
Optimized search ad fine-grained embedding models, boosting clustering advantage by +10.37%, NER F1 by +9.56%, expanding long-tail PV coverage by +3.98%, and achieving an 82.7% topic adoption rate.
NLP Algorithm Intern
Beijing, Beijing, China
→
Summary
Optimized full-chain document retrieval systems and developed context-aware sentence extraction methods, significantly improving retrieval accuracy and abstract quality.
Highlights
Optimized full-chain document retrieval systems, achieving a +4.2% increase in ROUGE-L F1 score for enhanced relevance and precision.
Improved embedding representation quality by generating and integrating summaries and keywords into document chunks, enhancing contextual understanding.
Developed a novel context-aware sentence-by-sentence abstract extraction method for Chinese conference abstracts, outperforming direct extraction by +12.0% in ROUGE-L F1 score.
Technical Research Intern
Shenzhen, Guangdong, China
→
Summary
Developed high-dimensional data visualization systems for MOBA game analytics and abnormal player detection, enhancing data analysis and system performance.
Highlights
Developed a high-dimensional data visualization system for MOBA game battlefield graphics and abnormal player detection, improving analytical insights.
Implemented distributed DBSCAN using Spark SQL, reducing distance calculation complexity and leveraging GraphX for efficient partition clustering.
Designed and developed hybrid Bezier spline-vector field visualization components using React and Fabric.js, enhancing interactive data representation.
→
Master
Information Resource Management (NLP Direction)
→
Bachelor
Information Management and Information System
Awarded By
CSDN
Recognized as a leading contributor of high-quality content in the Python programming community.
Awarded By
China University Computer Design Competition
Recognized for outstanding computer design and innovative project development.
Awarded By
China Collegiate Computer Competition
Recognized for outstanding mobile application innovation.
Awarded By
COMAP
Achieved finalist status in a prestigious international mathematical modeling competition.
Awarded By
Asia-Pacific Mathematical Contest in Modeling
Awarded for excellent performance in mathematical modeling.
Issued By
National Computer Rank Examination Committee
Issued By
IBM
Python, SQL, JavaScript, TypeScript.
Large Language Models (LLM), Natural Language Processing (NLP), Deep Learning, Algorithm Optimization, AIGC, DPO, SFT, RAG, CoT Prompting, Hallucination Detection, Embedding Models, Topic Modeling, NER, LLM Distillation, Machine Learning.
PyTorch, DeepSpeed, vLLM, ms-swift, FastAPI, Flask, React, Spark SQL, GraphX.
Elasticsearch, Qdrant, MongoDB, Distributed Systems.
Linux, Git, Docker, Full-stack Development, Backend Development, Frontend Development, System Optimization.
Data Visualization, High-dimensional Data Analysis, Statistical Modeling, Bayesian Inference, KPI Tracking, Data-driven Solutions.
Summary
Developed an intelligent platform for scientific research topic selection, integrating hot topic analysis, value assessment, and AI-driven generation, recognized with a National Second Prize in the China University Computer Design Competition.
→
Summary
Led comprehensive research on LLM hallucination detection, systematically evaluating 8 methods and analyzing how CoT prompting complicates the identification of hallucination cues.
→
Summary
Modeled and analyzed Wordle win rates, proposing a Letter2Vec solution for sparse letter representation and applying Bayesian inference with posterior probability correction to optimize game step decomposition.