Ryan Strauss

I’m an Applied Scientist at Amazon working on LLM post-training and routing for Alexa for Shopping (formerly Rufus). My interests and experience span language models, generative modeling, reinforcement learning, and learning from incomplete information. I enjoy bringing those ideas into real products, where model quality, careful evaluation, and the challenges of working at scale all matter.

Earlier at Amazon, I worked in Sponsored Products on forecasting, uncertainty estimation, and sequential decision-making. Alongside the modeling, I built data pipelines processing hundreds of terabytes and evaluated models through A/B tests. That experience continues to shape how I approach applied ML: thinking about the whole path from a research idea to a system people use.

Before Amazon, I was a graduate student at UNC Chapel Hill working with Junier Oliva in the LUPA Lab. My research focused on arbitrary conditioning—making predictions from whichever pieces of information are available. I developed methods for this across variational autoencoders, Transformers, and energy-based models.

As an undergraduate at Davidson College, I worked with Professors Raghu Ramanujan, Michelle Kuchera, Tabitha Peck, and Bryce Wiedenbeck on projects ranging from deep learning for nuclear physics to reinforcement learning for redirected walking in virtual reality.