At SIGMOD 2025, we demonstrated PalimpChat: a conversational interface for building optimized AI data pipelines.

I spent a week at Dagstuhl discussing the challenges and opportunities of table representation learning.

The DuckDB team used Pollock to test and improve its CSV reader on messy, real-world files.

We organized the first NOVAS workshop, dedicated to novel optimizations for visionary AI systems, at SIGMOD 2025.

On my visit to TU Berlin to talk about data accessibility and LLM-powered systems.