sqpe.dev Contribute
← Back to research
2026 / Optimization

CADENZA

CADENZA: Compiling Natural-Language Intent into Task-Specific Operator DAGs for Semantic Query Processing

What it contributes

Compiles a semantic operator’s natural-language intent into typed task DAGs. Task-extended relational algebra exposes intermediate outputs for rewrites, while physical planning jointly tunes backend routing and execution parameters.

How to read the evidence

Alternative task decompositions can change quality. Reported trade-offs depend on validation workloads, candidate plans and optimization overhead; they are not universal semantic equivalences.

We checked the primary publication record and abstract. This profile summarizes author-reported work; it does not claim a full-paper audit, implementation audit, or independent reproduction.

Explore the connections

Query optimizationMultimodalModel routing
  • UQE — UQL / SQL dialect
  • SemBench — Benchmark workloads
  • iPDB — Extended SQL