Hi! My name is Carlos Oliveira and this is my personal website. I am a Machine Learning Engineer with a training background in Experimental Particle Physics.
I grew up within a family of subsistance farmers, in Cacia, a small town of the “county” of Aveiro, Portugal. I worked part-time as a construction painter so I could afford my studies after middle school. It was a difficult but very rewarding path!
I graduated with a BsC in Physics degree in 2007 from the University of Aveiro, Portugal.
thesis, with the title “Monte Carlo study of electroluminescence in gaseous detectors”. My dissertation work consisted of using and extending the Garfield ++ and Magboltz platforms to simulate the electroluminescence produced during electron drift in noble gases. A few examples of my work are available here. Detectors based in this process have application in multiple areas: Medical Imaging , Energy Dispersive X-ray Fluorescence (EDXRF) imaging, High Energy Physics and Dark Matter Research. My work was supervised by Prof. Dr. Joao Veloso, Prof. Dr. Antonio Luis Ferreira and Rob Veenhof.
I spent the last 2 months of 2011 in East Timor teaching Physics to Timorese teachers, within the scope of a Bachelor Studies Plan to build up the Timorese high school system. It was an AMAZING and extremely rewarding experience!
I moved to Berkeley, CA, USA in early 2012 and joined the Lawrence Berkeley National Lab - Physics Division as a Post-Doctoral Fellow. There, I studied xenon additives (e.g. tri-methyl-amine) for improved performance in terms of energy resolution (important for Neutrino Less Double Beta Decay searches), and aiming the sensing of Weakly Interacting Massive Particles (WIMPs) induced nuclear recoils directionality in monolithic massive detectors, specifically Time Projection Chambers (TPCs). I was lucky enough to have the opportunity of working under direct supervision of the inventor of the TPC himself, Dr. David Nygren. This work was embedded in the Neutrino Experiment with a Xenon TPC (NEXT) international collaboration, of which I was a member between 2010 and 2015.
Between 2008 and 2014, I worked in the simulation and optimization of Micro-Pattern Gaseous Detectors like Micro-Hole & Strip Plate (MHSP), Gas Electron Multiplier (GEM), Thick-MHSP, Thick-GEM and Micromegas, as a member of the CERN RD51 Collaboration . I served as co-convenor of Working Group 4 (Simulations) between 2011 and 2014.
In 2014 I made a big bet, transitioning from the field of Physics to the intersection of Machine Learning and Bioinformatics. I joined, initially as a Scientist and then as the Director of Data Analytics, the Biodesix, Inc Research and Developement Department. Working in the beautifull town of Steamboat Springs, CO, USA, we combined Matrix-assisted Laser Desorption / Ionization - Time of Flight spectrometry (MALDI-TOF) with deep learning inspired algorithms to create new blood-based diagnostics for cancer. I contributed for the development of new algorithms (published in peer-reviewed journals) to address the curse of dimensionality that strongly conditions the applicability of traditional Machine Learning techniques in the clinical setting. Using such algorithms, we created generalizable and clinically useful cancer diagnostics, some of them protected by issued patents.
In late 2020, I joined the Intuit Artificial Inteligence team as a Staff Data Scientist where I developed Expert Productivity solutions leveraging Natural Language Processing and traditional AI. Namely, I worked on extractive and abstractive, Transformers based, text summarization for customer center calls: the team that I was part of was awarded the 2021 Scott Cook Innovation award for this work. I also used gradient boosting to guarantee tax return quality standards to power scalibility of the Turbotax Full Service offering. While working for Intuit, in 2021 I made the decision to move to San Diego, CA, USA.
Recently, in July 2022, I joined Twilio as a Principal Machine Learning Engineer. There, I’m a Tech Lead for Predictive Traits - a scalable, customer facing product that allows marketers to easily create multivariate, machine learning models to predict user propensity to perform an arbitrary action. The product allows marketers to leverage advanced ML techniques without the need for deep technical knowledge, with just a few clicks, and using the user data they already track through Segment.
- About me
- In the news