Principal Machine Learning Scientist
Pedro Tabacof
Principal Machine Learning Scientist at Fin · Ireland
At Fin, I lead Fin Core, a 12+ person workstream that defines how the AI agent answers questions, follows guidance, takes actions, and handles the conversation around them.
In more than ten years, my models have informed a $10M annual marketing budget at a mobile gaming company, credit decisions involving tens of millions of dollars in exposure at Nubank, and fraud screening for more than two-thirds of iFood transactions. My research includes an ICLR main-conference paper and four NeurIPS workshop papers, with 600+ citations.
Email · LinkedIn · Google Scholar · GitHub
Recent posts
2024
How I lost 1000€ betting on CS:GO — Practice
What happened when the theory met feature engineering, validation, backtesting, and real money.
2023
Name classification with ChatGPT
Comparing prompt-based classification with traditional language models.
2023
Real-Time ML Models with Serverless AWS
A practical guide to deploying models with AWS Lambda and API Gateway.
2023
The Hierarchy of Machine Learning Needs
A framework for deciding what matters before choosing or tuning a model.
Elsewhere
Why not just ship it?
How we evaluate changes to Fin before and after release. 2026.
Do Small Language Models Learn to Negotiate?
A controlled study of reinforcement-learning-trained sellers. NeurIPS 2026 SLM-Agents workshop (poster).
Slower Feels Smarter?
An experiment on latency and user perception in AI agents. 2025.