- Leading the engineering team in charge of developing and building DeepMind's AI Data Platform and its underlying organization to facilitate the development of SOTA GenAI models such as Gemini and Gemma.- Leading and coordinating initiatives in Data Governance, Data Quality, Safety and Infrastructure for all GenAI projects within the company.- Leading the GDM DMA program and coordinating all related efforts at GDM.

Pioneering the DataPrepOps (a term I coined in January 2021) space and using MLOps + AI to automatically prepare, curate, tune and optimize ML training datasets for specific use cases and models.

Currently advising: Mimoto AI, Mpathic, inGen Dynamics. Past companies: Veamly, VoiceMe, among others.

- Educated operators and executives from Fortune 500 companies on AI, Data and MLOps.- Advised companies to help them build world-class ML teams and strategies.

Figure Eight (aka CrowdFlower) was the first company to offer enterprise-grade labeling and human data solutions to ML teams. I built and grew the team and technology that led the company to a $300M acquisition by Appen within a year of my hire. I left right prior to the acquisition to pursue my entrepreneurship ambitions.

Built Atlassian's first centralized Machine Learning function from scratch and ran the Search & Smarts team, a cross-functional team in charge of developing all Machine Learning-powered features for all Atlassian products.Grew a world-class team of data scientists and engineers from 3 to 30+ people in less than 6 months, across 3 locations (Mountain View, San Francisco and Sydney).Products developed by my team:- Cross-Product Search: Centralized Search Engine to search across all Atlassian products- User Search, Smart Mentions, Recommendations, News Feed

My achievements as a leader of the Search Relevance team- Build and led the Metrics Measurements & Insights team, a group of 12 people (including product managers, engineers and data scientists), in charge of advancing the usage of data and ML best practices within Search.- In charge of all Human Evaluation initiatives / preparing training data for the rest of the department.- Bootstrapped the Store-Search team, the first ever group data scientists allowed to operate across Walmart stores and Walmart eCommerce. The iniative broke down one of the trickiest silos at Walmart and aimed at democratizing the usage of data from Walmart Stores to improve the algorithms used on Walmart.comTeam projects:- Metrics development: our team developed key metrics for the organization, such as item discoverability, customer satisfaction and LTV, diversity and perception indexes, using both advanced analytics strategies and ML.- Machine Learning Lifecycle Management / Automated Model Management: the team developed tools to enforce and automate ML best practices and manage models without the need for human intervention and accounting for systems performance and cost.- Browse Ranking: the team's new solution increased the revenue coming from browsing customers by ~10%.- Human Evaluation: our team created active learning algorithms that cut in two the volume of data that the Search Department originally had to be evaluated manually.- Stores-Search: the team developed multiple signals and indicators based on Walmart Stores data to improved the accuracy of the company's Learning-to-Rank models.

- Review Analysis: I invented an advanced strategy using Post-Purchase Customer Feedback to advise customer purchases (3 patents covering the technology).- Various projects to advice the Data Strategy of the Department and guide Product Development.

Ayasdi (now known as SymphonyAI) was a VC-backed Machine Intelligence company that developed the concept of Topological Data Analysis (TDA) as an innovative way to complement traditional statistical techniques for data analysis and machine learning.As a Principal Data Scientist, I worked both as on pushing the boundaries of the software and on researching new ways to leverage its capabilities when solving real-life problems.- Worked on all kinds of business problems, from looking for outliers to analyzing customer behavior. The industries I've worked with range from retail, social media to cyber security and the financial industry.- Developed code to automate data preprocessing and investigated data fusion solutions.- Actively contributed to the design and development of novel algorithms in Topoligical Data Analysis

Leadership/managerial experience:As the founding member of the Data Science and Engineering team (which I grew to 8 people in 9 months), I lead all data initiatives, built the roadmaps and acted as the team’s spokesperson. I also worked closely with the executive team on collecting more data features to increase the diversity of the company's data, and build a stable data function.Technical experience:My responsibilities covered a large range of Data Science problems, such as forecasting, supply allocation, fraud detection, customer acquisition, etc. Here are some of the projects I worked on:- Design of multiple algorithms to classify our publisher inventory, based on NLP (Bag-of-words and n-gram models)- Prototyping, development and deployment of the company's fraud detection system.- Prototyping, engineering and deployment of YuMe’s inventory forecasting algorithms using Latent Class Modeling.- Development of a scalable supply allocation system using an hierarchical clustering approach.- Others: user and family-level fingerprinting, design of attention and receptivity classifiers, targeting tools.

My job at Quantlab Financial consisted in maintaining and enhancing fully automated market microstructure models and trading algorithms, and assisting the traders in the choice of parameters/indicators to be used in their daily strategies. This involved both development of new algorithms and maintenance of existing code and strategies.My most exciting project was the conception, prototyping and development from a range of NLP algorithms aiming at forecasting short-term market price movements by analyzing information found in news articles and social media (using sentiment analysis and topic modeling).

Engineering experience:Worked on Monte Carlo Simulation and Software Development for the needs of the LBNE experiment, a neutrino experiment to be built in South Dakota. I focused on the design of LBNE's to-be neutrino detector and worked hand-in-hand with engineers to bring to life its blueprint with, as a main constraint, the budget that was allocated to the project by several governmental agencies.Analytical experience:As a member of the Super-K Collaboration (Japan), I also worked on several Data Analysis projects, and also contributed to an effort consisting in translating their old software (written in Fortran) into C/C++ and robustifying it.

Particle Physics experiments are typically run by large collaborations involving 500 to 1,000 people from all backgrounds: hardware engineers, computer scientists, physicists, etc.As an Experimental Particle Physics, I had the chance to develop simultaneously my math and analytical proficiency, my engineering expertise and early leadership skills.While I was a Research Assistant at SLAC I studied CP-violation in the decay channel of a particle called a B meson into a Ks, a charged pion and a neutral pion.Some of my responsibilities as a member of the BaBar collaboration:- My main work consisted in the complex data analysis of a decay channel never attempted before, using a "Dalitz analysis" process, a maximum likelihood estimation fit as well as other supervised machine learning techniques to reduce the amount of background noise. This specific decay was the B+/- --> Ks pi+/- pi0 3-body charmless B decay channel.- As a 'Package coordinator', I was in charge of the maintenance/support/development of software used across the collaboration.- As a DIRC data quality expert, I was in charge of the quality of DIRC subsystem data taking process (looking for abnormalities in the data, and taking away outliers due to a defect in the system)- As the 'DIRC commissioner' of the BaBar detector, I led data collection efforts and other projects aiming at improving the overall efficiency of the DIRC system, and was in charge of its maintenance.Keywords: Machine Learning, Data Mining, Noisy Datasets, High-Dimensionality Datasets, Monte Carlo Simulation, Artificial Neural Networks, Random Forests, Maximum Likelihood Estimation, Clustering, Sampling

I taught:- Newtonian physics and hydrodynamics (Sep 2008 - Jan 2009)- Mathematics for physicists (Sep 2007 - Jan 2008)- Particle & nuclear physics for physics students (Feb 2007 - Jun 2007)- Biostatistics for medical students (Sep 2006 - Jan 2007)I gave lectures, conducted experimental classes, graded papers and wrote examination subjects, among other responsabilities.

I worked with the BaBar team in Paris and contributed to research efforts on the B0 -> K+ pi- pi0 and B0 -> Ks pi+ pi- decay channels.

I worked on the development of the trigger system for the Double Chooz Experiment. My teammate and I contributed to the Electrical Engineering part as well as the Software Engineering of the drivers of the system.

I interned with the D0 team and studied the decay of the Higgs into W + jets channels. D0 is a collaboration of about 500 physicists hosted by the TeVatron/Fermilab near Chicago, IL. My main responsibilities included the software development of tools as well as MC simulation work.