-
Text Compression Through the Looking Glass
Cold and unstructured text has long been a storage burden, driving costs for data that is unlikely to ever be accessed again. The rise of accessible large language models (LLMs) has intensified this challenge by dramatically increasing the volume of generated content that still needs to be retained, e.g. for compliance reasons. This post explores a new class of LLM-based compression methods that can significantly reduce the storage footprint of text-heavy data, and explains why LLMs are particularly well-suited to text compression.
-
String Fingerprints
Cloud data warehouses are text-heavy. As the amount of text data to scan increases, queries become slower, therefore query engines require fast pre-filters to accelerate them. We present string fingerprints, a lightweight secondary index structure designed to approximate LIKE predicates, albeit with false positives. Fingerprints can be optimized for specific workloads using mixed-integer optimization and even generalize to unseen table filters.
-
Benchmarking Semantic Query Processing Systems
Semantic query processing is emerging as a new layer atop relational engines, elevating LLM-backed semantic operators to first-class SQL primitives for multimodal data. We present SemBench, the first benchmark to rigorously evaluate these systems end-to-end, and outline our roadmap towards our own system, Spectra, to make semantic operators affordable at scale.
-
Democratizing Data Science
Our vision is to build an end-to-end agentic data platform, enabling domain experts to acquire, clean, analyze, and visualize data in a principled manner by combining the benefits of LLMs with decades of database research.
-
Launching Our Blog And Wrapping Up 2025
I'm super excited to launch our blog! We'll use this space to share what's happening in our lab, from research papers and systems to the day-to-day life of our team. To kick things off, let's look back at 2025.
We use analytics cookies to understand site usage.