Vivien Xie

Senior Data Scientist Machine Learning Engineer @LexisNexis

New York, NY, US

Contributed to designing and developing Generative AI applications that leverage Large Language Models (LLM) to power legal research and legal drafting services. • Formulated contextually grounded prompts by integrating structured templates and incorporating passages from the internal Question and Answer retrieval system (Retrieval- Augmented Generation), which improved response usefulness by 25% and reduced hallucinations by 6%. • Implemented asynchronous method calls to internal services, leading to a 53% reduction in response time. • Contributed to designing and implementing Multi-turn Conversational Strategy. • Preprocessed and generated data from a diverse legal corpus across multiple international jurisdictions for model training. • Fine-tuned a Sentence-Transformer Embeddings Model on legal contents for embedding search. • Performed model evaluations on OpenSearch indices and achieved nDCG@1,3,5 increases by 4.82, 3.26, 3.59; Precision@1,3,5 increases by 4.71, 2.35, 2.38; Recall@1,3,5 increases by –2.24, 2.47, 3.9. Reduced operational cost by 30%. • Experimented with various open-source Large Language Models (LLM), evaluating the time and budget implications for causal language model fine-tuning and instruction fine-tuning using a curated legal corpus. • Collaborated on automating legal citation recognition using Named Entity Recognition (NER) and Entity Linking algorithms, experimenting with spaCy, Flair, BERT models. Achieved final Precision 95.26%, Recall 96.08% and F1 Score 95.67% with Flair model, reducing document processing time and editing cost by 50%. • Designed metadata schema and published metadata in the DataLake. • Actively contributed to the product\'s release cycle and verified the accuracy and quality of its outputs. • Participated in technical conferences, presentations and evaluated new technologies.

New York University

Master of Science (MSc), Financial Engineering

University of Pennsylvania

Master's degree, Computer and Information Technology, 3.93