Tong Tony Lai

Bachelor's

Business

Computer Science

Microsoft

SOFTWARE ENGiNEER INTERN

• Developed two novel training pipelines that leveraged SSL pretraining for encoder-only deep learning models; improving downstream task f1 performances on Entity Matching by an average of 2% for benchmark datasets and 4% for internal datasets. Built with Python using NumPy, Pandas, Hugging Face, Natural Language Toolkit (NLTK) libraries. • Productionized training pipelines’ updated models; refactoring our Entity Matching API infrastructure code to utilize a new transformer model using C#, yielding a 5% improvement in our ranking output metric for our measurement set. • Organized and led presentations on linguistic and deep learning concepts, presenting papers to the team to introduce pertinent best practices and innovations for NLP work