About

My name is Emmanuel, and I’ve worked in Machine Learning for years. I write to share the most important lessons I’ve learned over years of working in applied ML and research.

I’m currently at Anthropic on the interpretability team. My research is about understanding how LLMs like Claude work internally. Previously, I worked on the finetuning team focusing on tool use. Before joining Anthropic, I was at Stripe, where I made foundational improvements to core fraud models and led efforts to automate the training and deployment of ML models. Previously, I was Head of AI at Insight, where I helped hundreds of Fellows build applied ML projects. Before that, I worked as a Data Scientist at Zipcar and LocalMotion building data products.

Recent Research:

• When Models Manipulate Manifolds: The Geometry of a Counting Task
• Tracing Attention Computation Through Feature Interactions
• On the Biology of a Large Language Model
• Circuit Tracing: Revealing Computational Graphs in Language Models
• Interpretability: Understanding how AI models think

I’ve previously written articles on Medium, but have decided to move my writing to this blog. Join the mailing list below to be notified when I publish new articles and tutorials.

This website is powered by Hugo, using a slightly modified version of the Coder theme.