The Stretto Execution Engine for LLM-Augmented Data Systems
Gabriele Sanmartino, Matthias Urban, Paolo Papotti, Carsten Binnig · 2026
Read The Stretto Execution Engine for LLM-Augmented Data Systems by Gabriele Sanmartino, Matthias Urban, Paolo Papotti, Carsten Binnig (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04430.txt.
Description
Queries that mix database tables with language model judgment need an engine that budgets accuracy. Gabriele Sanmartino and 3 co-authors present Stretto, an execution engine for semantic data systems, in a 2026 paper. These operators can run in many ways, each trading speed against accuracy.
Unlike earlier systems, Stretto promises accuracy across the whole query plan, not per operator. Planning becomes constrained optimization: a gradient-based optimizer picks implementations and splits error budgets jointly. A KV-cache compression trick supplies a spectrum of execution strategies with controllable trade-offs. The design targets analysts querying structured and unstructured data together without uneven quality.
Catalog record
| CREATOR | Gabriele Sanmartino, Matthias Urban, Paolo Papotti, Carsten Binnig |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04430