MiguelMerlin
mmerlin@stevens.edu
CS & Math at Stevens
Institute of
Technology &
Software Engineer at
Amazon
Hello! I am Miguel. I love working on challenging and impactful problems. Currently working on physics informed neural networks, large language models, deep learning theory and software at scale.
I am a junior at Stevens Institute of Technology, pursuing degrees in Computer Science and Mathematics. At Stevens, I actively work on advancing machine learning methods across various domains.
I have worked with the Alexa organization at Amazon during the past two summers in Bellevue, WA. My contributions centered on developing an experimental framework to benchmark the performance of statistical models responsible for routing user utterances to the appropriate skills and service domains. This framework ensured traffic was diverted from underperforming models.
Last summer, I played a role in advancing the transition toward an LLM-powered Alexa, contributing to groundbreaking innovations in conversational AI. My primary focus was supporting the Chain-of-Thought (CoT) model, enhancing its reasoning capabilities to deliver more contextually relevant and coherent responses. This involved addressing complex engineering challenges such as optimizing inference latency, managing memory constraints, and ensuring the stability of multi-turn interactions in real-time production environments. Additionally, I worked on Retrieval-Augmented Generation (RAG) methods to further refine the model's ability to provide accurate and context-aware responses.
For the past year, I have been a member of the ScAI (Scientific AI) lab, exploring the intersection of numerical methods in physics and machine learning. I am collaborating with Los Alamos National Laboratory on a research project developing ML models to predict gas saturation diffusion in porous media. The computation of solutions to differential equations is compute intensive, therefore my research focuses on embedding physics-based constraints into the training process of Machine Learning models to Neural Operator Networks that can solve DEs with high accuracy, and low computational cost.
I actively serve as a VP of Technology at Blueprint, a student-run organization at Stevens that provides software solutions to non-profits at no cost. I am responsible for leading the development of software solutions for our organizations, managing the technology team, ensuring the quality of our deliverables, and managing our cloud infrastructure.
Please feel free to reach out to me at mmerlin[at]stevens[dot]edu. I am happy to chat.
News
Blog
Coming Soon!