Barcelona, Catalonia, Spain
My 12 years of experience have trained me to quickly provide new innovative solutions to complex problems but also to lead projects to improve the quality and efficiency of more old-school Data Science practices. In the past I have been a Research Scientist, with publications in Astrophysics but also experience at a ML Research industry group. I have also been a Data Scientist with experience at a pre-IPO company (Stitchfix) but also at more mature companies (Spotify, Netflix). My projects have included analytics, experimentation, Data Science tooling, statistical modeling and Machine Learning, using a wide variety of tools (Python, R, SQL, Spark, etc.).
Leader of a product area with about 50 ML, Backend and Data Engineers, responsible for strategic interventions within our main recommender systems. This includes long term recommendations, strategic promotions to enhance the reach of our content creators and cutting edge ML technologies such as explore/exploit and multi-objective balancing.
Leader of a product area in charge of several engineering squads, responsible for strategic implementation of long term behavioral models at scale across the company. This includes long term business decisioning, AB testing and algorithmic recommendations. I am also responsible for the organizational health of a team of 30.
Member of the Discovery ML Research team. - Contributed to an internal ML python package that leverages ensemble methods to forecast the survival rates of subscribers. - Led a project to implement novel techniques that use the outcomes of ML models to improve on the decisions made on A/B experiments. - Contributed to a reinforcement learning service that delivers personalized emails and notifications to users based on their behavior.
Member of the Product Data Science team. - Analyzed and helped design dozens of product UI experiments to improve the user experience. - Studied historic experimentation records to improve A/B testing methods and define better metrics that capture user behavior. - Created and productized a statistical model that predicts the quality of a network connection based on app performance data, and used it for experimentation segmentation. - Led development of a project to leverage machine learning models on mobile applications.
Member of the Merchandising algorithms team, which is in charge of building recommendations to improve the quality and assortment of Stitch Fix inventory. - Developed tools to identify temporal and seasonal trends from client feedback. - Built recommender systems for fashion design. - Created ETL pipelines and reporting tools, and migrated old ones to Spark.
Implemented algorithm in Python using the scikit-learn package that transforms the detection of short-period planets in time-series data into a classification problem, where features are measured from the Fourier transform of the data. Implementations with Support Vector Machines achieve 99% accuracy on simulated data. (https://github.com/sanchisojeda/FFTMLproject) Designed a framework to acquire raw data from the K2 telescope, clean it, analyze it with statistical tools, and share the clean data with collaborators. Founder and manager of the international research collaboration ESPRINT to discover rocky planets using data from the Kepler space telescope. Designed platform to acquire and process raw data from the trimestral public releases of the Kepler K2 mission. Shared clean data with collaborators, and analyzed it using time-series analysis tools, maximum likelihood models and Markov Chain Monte Carlo routines. Wrote publicly available suite of Python tools that use Numpy, Pandas and Matplotlib to classify an exoplanet as rocky by measuring its mass and radius from time series data. (https://github.com/sanchisojeda/USPmass) Designed and supervised data-intensive computational research projects for five undergraduate and graduate students.
Created tools to plan ground based observations and to measure the masses of close-in rocky planets.
Created new technique and algorithm to find short-period planets. Millions of simulated signals were used to benchmark the technique, proving an order of magnitude computational speed increase over existing techniques. Designed a new algorithm to measure the obliquity of an exoplanet host stars using archival data. The algorithm was successfully applied it to several complex exoplanet systems.