About Me

“Generating numbers is easy, generating numbers you should trust is hard!”

Broadly-trained data scientist with 7 years of experience, accomplished Data Scientist in complex experimentation/causality, time series forecasting, and classification models. My background is statistics with lots of side studies on economics

I have worked as a Research Assistant for a Brazilian National Research Institute, a Statistician at Labour Ministry, Product Analyst at the largest media company in Latin America (Globo), Marketing Scientist at Uber and as Marketing Data Scientist at Glovo. Currently I’m working for Toptal as Strategy Analyst. My main areas of interest are causal inference, econometrics, and machine learning.

Proven business impact in different areas, particularly Marketing, Experimentation and Operations in the last years. Skilled at optimal decision making, causal inference, machine learning, communication, stakeholder management, business acumen, OKR’s setting and tracking and problem-solving. I’m driven by curiosity, challenging problems, chewing technical and academic books and seeing a problem from multiple perspectives.

Experience

@Toptal

Strategic advisor for Toptal business functions and leaders
Leading Talent Operations analytical function.
Holding weekly analysis prioritization meetings
Holding monthly business performance review meetings
OKRs definition
Improved newcomers talents retention by 5% in the network and by 2% the talents allocated to a job within 10 days. Discovery
- Discovery - discovered talent retention issue. The metric wasn’t being tracked.
- Problem Statement - explored the data and found that many talents leave the network after finishing a job.
- Action Item - Implement a new job matching process to tackle jobs with end scheduled.
Probability to convert based on Technical Steps Grade - Regression Model to predict when a talent score high in the first steps of the funnel if we can skip further steps in the screening process.

@Glovo

Experimentation/Causality POC
Identifying heterogeneous treatment effects of marketing interventions (Retargeting). Segment Discovery leads to more efficient retargeting audiences.
Multi-Touch Attribution (building the role pipeline for u-shaped and Markov chains)
UMAP embedding clustering to identify users that churned due to bad user experience.
Marketing First Order Prediction - predicting total conversions for a specific time window.
Incrementality Tests for Retargeting and Acquisition
Build reliable, precise and faster A/B tests with variance reduction (CUPED), bias correction (CUPED / Diff and Diff), identifying novelty or primacy effects, randomization unit correction (Linear Mixed Model), and Bayesian Inference.
Moving the company culture of decision making based on correlation studies to causality analysis (instrumental variables, propensity score matching, mediation modeling, regression discontinuity, causal diagrams, and others causal inference methods)

@Uber

Methods for geo experimentation: Bayesian Hierarchical Time Series (Causal Impact), Synthetic Control, Matching Similar Cities and Regions for geo experimentation
Market Saturation, Total Addressable Market and Opportunity Size for Engagement, Churn and Acquisition segments
Marketing and Incentives Impact in Acquisition Metrics - short and long term impact inference.
LAtam POC for Marketing Incrementality Tests & Measurement Methodology
Measuring Offline and Brand Campaigns Business Impact
Marketing Business Results Report sent monthly for key stakeholders
Interviewing marketing data analysts

@Globo.com - largest media company in Latin America

Daily business insights giving guidance for product changes (working along a software engineer and a designer)
Implemented a churn prediction model to anticipate this decision and make actions to retain the customer.
User Path Navigation with Markov Chain
Building the metrics and the statistical environment on A/B platform.
Formulating success metrics for quality of experience on player, estimating marginal effects of these metrics on engagement and using the estimates on how to prioritize what metric to act on.
Computer Vision for predicting when the video credits start

@Ministry of Labor

Time series prediction of admissions and dismissals for the main Brazilian cities
Monthly report of labor market monitoring metrics of affiliated municipalities
Public Dashboard informing policy makers and the overall population in the current labor market situation

@IPEA - Institute of Economic and Applied Research

An econometric analysis of the diversity on agriculture familiar production.
Panel of social vulnerability in partnership with the UN.
Investigates the population and employment dynamics in central urban areas of twelve selected capitals using Spatial analysis methods. Kernel heat maps were used to conceptualize and delimite central areas for each city.

Skills

Favorite machine learning frameworks: Scikit-learn, Caret, XGBoost, LightGBM, Keras
Programming Languages: R, Python, SQL
Compute Instance: Google Compute Engine
Data Warehouse (OLAP): Hive, Vertica, Presto, AWS Redshift, Google BigQuery
Cloud Storage: AWS S3, Google Cloud Storage
Distributed Machine Learning: Spark
Job Scheduler: Luigi
Visualisation: Tableau, Matplotlib, ggplot, Looker, Google Spreadsheet
Version Control: Git
Project Tracking: Jira, Trello

Courses

MITx: 15.071x The Analytics Edge: MIT’s The Analytics Edge is an edX course focused on using statistical tools to gain insight about data and make predictions. It has around 75 datasets and starts from linear regression upto clustering and some classification techniques like Random Forest and CART models in between.
Data Science Specialization: This Specialization covers the concepts and tools you’ll need throughout the entire data science pipeline, from asking the right kinds of questions to making inferences and publishing results.
Master Statistics with R: This Specialization showed how to analyze and visualize data in R and created reproducible data analysis reports, demonstrate a conceptual understanding of the unified nature of statistical inference, perform frequentist and Bayesian statistical inference and modeling to understand natural phenomena and make data-based decisions.
Digital Marketing: This program offers you the opportunity to master platform-specific skills valued by top employers, while at the same time establishing a broad-based understanding of the whole digital marketing ecosystem. Run live campaigns on major marketing platforms, learn and apply new marketing techniques, analyze results, and produce actionable insights.
Causal Inference: In this series of 7 courses created by Duke University with support from eBay focuses on how to use the advanced methods of instrumental variables and regression discontinuity to find causal effects. Guides you through the concepts that data scientists need to always consider when examining and making inferences about data.

Publications

Rabetti, M.S.; Nadalin, V.G.; Oliveira, C.A.P.; Furtado, B.A.; Cavalcanti, C.B. (2016) Population and Employment Dynamics in the Urban Centers of the Brazilian Metropolis .
Rabetti, M.S.; Sambuichi, R.H.R.; Galindo, E.P.; Pereira, R.M.; Cconstantino, M. (2016) Production Diversity in Family Agriculture Establishments in Brazil: an econometric analysis based on the registration of the Declaration of Aptitude to Pronaf (DAP) .
Rabetti, M.S. and Carvalho, C.H.R. (2015) Traffic accidents on Brazilian federal highways .

About This Site

This site is powered by Jekyll using the Minimal Mistakes theme. All blog posts are released under a Creative Commons Attribution-ShareAlike 4.0 International License.

All R blog posts are compiled with knitr R markdown. You can find the reproducible sources of each blog post here.

Matheus Rabetti

Recommended Blogs