Vincent Lin Email & Phone Number

Cambridge, Ma, Us

Cambridge, Massachusetts, United States

- Designed, trained, and evaluated a novel multi-modal large language model architecture for zero-shot cell type annotation by adapting a vision-language model architecture for single-cell genomics & transcriptomics applications- Achieved state-of-the-art performance in zero-shot cell type annotation, with plans to publish results in a conference paper- Optimized model training speed & memory consumption using JAX and Haiku, which involved model sharding, data parallelization, memory allocation management, and multi-TPU optimization to scale architecture from a small language model to a modern LLM

Seattle, Washington, United States

- Evaluated five multi-agent large language model frameworks from academic literature for quantitative & qualitative performance across various tasks, including persona simulation, code generation, modular device design, and simulated human interaction- Prototyped models integrating multiple multi-agent frameworks in PyTorch for LLM-simulated human conversation, personality-driven decision making, and environment manipulation via multi-LLM interaction, LLM planning/reasoning, and prompt engineering- Demonstrated & communicated the capabilities & limitations of customer-centric multi-LLM generative AI to Amazon stakeholders

Cambridge, Massachusetts, United States

- Conducted research related to data privacy in machine learning, in the context of using an obfuscator model in an adversarial representation learning architecture to both anonymize data and produce mathematical guarantees on privatization- Pioneered work for developing an obfuscator architecture applicable in the context of modern transformer models to anonymize textual data for various natural language processing tasks- Extensive development in Python and PyTorch for various deep learning architectures

Cambridge, Massachusetts, United States

- Performed research related to data privacy in machine learning, in the context of using­ an obfuscator model in an adversarial representation learning architecture to both anonymize data and produce mathematical guarantees on privatization- Pioneered work for developing an obfuscator architecture applicable in the context of modern transformer models to anonymize textual data for various natural language processing tasks- Extensive development in PyTorch for various deep learning architectures

Omaha, Nebraska, United States

- Developed an integrated graph-based web application under a government military contractor used by the Air Force for communications analysis and warfare strategizing - Worked in a full-stack software environment in Angular frontend, Spring backend, and Elasticsearch database - Enhanced algorithms used for path analysis within the application via runtime optimizations, improved algorithmic design, and general documentation - Researched and prototyped implementations of graph convolutional neural networks in TensorFlow as a demonstration of ML/AI for use in the application data network- Collaborated with engineers in Agile development cycle, presented summer project to company executives at end-of-season conference

Cambridge, Massachusetts, United States

Worked with the InfoLab at MIT CSAIL to develop and improve the START (SynTactic Analysis using Reversible Transformations) Natural Language System, a search engine that utilizes natural language processing. - Designed scripts in Scheme and Bash for web scraping of around 60 online databases, backend interfacing, and file management- Maintained lab's PostgreSQL backend by writing SQL scripts to restructure data and performing administrative troubleshooting- Communicated issues, brainstormed solutions, and established goals at weekly standup meetings

Bachelor'S Degree, Computer Science

Activities and Societies: Undergraduate Research Opportunities Program (UROP), Undergraduate Practice Opportunities Program (UPOP), MIT.

High School Diploma

Activities and Societies: NAfME All-National Symphony Orchestra, National Honor Society, Tri-M Music Honor Society, Marching Band, Jazz.