The first NOVAS workshop at SIGMOD 2025

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

Why NOVAS

AI is becoming part of the data processing stack. Systems now combine relational operators with language models, multimodal models, vector search, and agentic components. This opens up exciting research questions around optimizing their accuracy, cost, latency, scalability, and reliability.

We created NOVAS as a place to discuss these problems across the data management and AI communities. For the first edition, we invited work on systems for large heterogeneous data collections, semantic operators, multimodal question answering, and database-inspired techniques for optimizing AI workloads. We accepted 5 papers for presentation at the workshop, with an ~80% acceptance rate.

A great first edition

After several months of calls, reviews, scheduling, and the many small tasks that hide behind the word organizing, the day finally arrived to host the workshop at SIGMOD 2025 in Berlin. We featured two keynotes, one by Carsten Binnig and one by Aditya Parameswaran, and it was a great pleasure to see a full room for the attendance of the workshop!

Many thanks to the speakers, authors, reviewers, attendees, and a special shout-out to my fellow organizers: Chunwei Liu, Lei Cao, Huan Sun, and Paolo Papotti. Hearing all the excitement around this area made the organizational work very much worth it :) Who knows, maybe we will see you at NOVAS 2026?!