Luke de Oliveira | Artificial Intelligence, Machine Learning

About.

I'm currently at Meta working on AI research (distillation, quantisation, pruning, etc.) for LLMs supporting the next generation of Llama. Previously, I was the Director of Machine Learning at Twilio, where I led the AI organisation. Before that, I was CEO & Founder of Vai Technologies, which was acquired by Twilio.

I am an active angel investor, and am particularly drawn to early-stage startups that have one (or more) of the following traits: AI-focused, API-focused (including all flavours of developer-first GTM motions), or B2B-centric. Whether or not I am an investor, I enjoy helping founders where I can.

For many years, I was in academia working at the intersection of Deep Learning and High Energy Particle Physics. I held appointments at Lawrence Berkeley National Laboratory (LBNL), SLAC, and CERN. A somewhat complete list of publications can be found on my Google Scholar.

I earned a B.S. in Applied Mathematics from Yale University and an M.S. in Computational Mathematics from Stanford University.

Organizational services

We help you determine the organization infrastructure necessary to build your products.

SERVICES one

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius.

Learn more

SERVICES two

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius.

Learn more

Interests.

A subset of things I like to build, tinker with, and research

Deep Learning

I design novel deep architectures, particularly for NLP and sometime CV. I'm interested in end-to-end learning from characters. I also like thinking about cloud deployment strategies for DL. I like Keras, but sometimes I pretend I understand TF.

Recommender Systems

I spend a lot of time thinking about how to make recommender systems better. I've been experimenting with some end-to-end approaches, but I mostly spend time automating subsets of the collaborative + content driven pipeline

Infrastructure

Machine learning is great in a vacuum, but it looses real-world oomf when your brilliant models are confined to an Jupyter Notebook. I use Docker + Vagrant for my deployments.

Predictive Analytics

More generically, we extract insights from data that you've collected. Even if you don't have a specific goal in mind, we can help you find insights you never knew were there.