Beyond Auto-Code Gen: Building Production-Grade Data Science Agents with Tabular Foundation Models

When autonomous AI agents are tasked with tabular prediction problems, whether forecasting sales, predicting churn, or detecting fraud, they almost always follow a flawed playbook. They write custom machine learning scripts, train models from scratch, hallucinate metric improvements, and force developers to manage complex model-serving pipelines for every single run. This talk introduces a paradigm shift for tabular machine learning in agentic workflows: Tabular Foundation Models (TFMs). By leveraging TabPFN (Tabular Prior-Data Fitted Networks), AI agents can move from fragile, code-generating ML loops to instant, deterministic in-context learning. TabPFN delivers state-of-the-art predictions in seconds; requiring zero ML code generation, zero model training, and zero artifact storage.
Format: TECHNICAL TALK, BUSINESS TALK
Location: ROOM A
Speakers: Eliott Kalfon
Level: Beginner
Audience: Registered participants