Experience

Professional experience and selected applied work by Pedro Tabacof.

Experience

Ten years of applied machine learning across AI products, games, credit, fraud, marketing, and research.

Fin / Intercom

Principal Machine Learning Scientist, previously Staff · 2023–present · Ireland

I lead Fin Core, the 12+ person workstream responsible for most behaviour between end users and Fin’s AI agent: answering informational and personalised queries, following customer guidance, taking actions, and handling conversational replies. I am responsible for strategy, roadmap, and execution.

I have managed three senior ML scientists across locations, including hiring and onboarding, and currently manage the lead of the Apex workstream.

Over the period I have led the workstream, Fin AI Agent ARR grew more than 30× and its resolution rate doubled—results delivered by the broader team. My individual-contributor work includes post-training open-weight models and migrating Fin to Anthropic models, spanning prompt engineering, evaluation, A/B testing, and cross-team delivery.

Public descriptions of this work:

Wildlife Studios

Senior / Staff Data Scientist and Data Science Manager · 2019–2023

I led the data-science work for a privacy-preserving iOS marketing-attribution system after Apple’s tracking changes. Its outputs informed roughly $10M in annual performance-marketing spend. Earlier, I built lifetime-value models for games including Zooba and Tennis Clash, which together passed 100 million downloads.

I also managed three senior data scientists and promoted one from senior to staff.

Nubank

Senior Data Scientist · 2018–2019

I built and deployed credit-risk and severity models, along with a net-present-value framework that connected model predictions to lending decisions. The work had impact measured in millions of dollars of profit and tens of millions in credit exposure. I also contributed to the open-source fklearn library.

iFood

Senior Data Scientist · 2017–2018

I developed a real-time credit-card antifraud model used on more than two-thirds of transactions, with estimated savings in the hundreds of thousands of dollars per year.

Accenture and academic research

Research Engineer · 2019

At Accenture, I worked on knowledge-graph embeddings, counterfactual explanations, and Bayesian deep learning. That work produced an ICLR 2020 main-conference paper on probability calibration and contributions to AmpliGraph.

My academic foundation is an M.Sc. from the University of Campinas (Unicamp), where I studied adversarial examples and uncertainty in deep neural networks.

Patent applications

These are published US patent applications, not granted patents.