VeriPIE: Checkable Query Plans and Evidence-Guided Selection for Text-to-SQL
We present VeriPIE, a neurosymbolic framework combining language-model generation with symbolic checks.
ProofPaper ↗
Key points
- Financial Artificial Intelligence (AI) governance and model risk management (MRM) emphasize independent validation, transparency, and effective challenge of AI-generated computations.
- Text-to-SQL translates natural-language questions into Structured Query Language (SQL), but queries that agree on one database can still express different computations.
- It ensembles direct SQL generation with QuerySketch, a structured plan language checked by a backend implemented in Lean, a theorem prover and programming language.
- Plan diagnostics guide repair before SQL generation, while execution agreement and bounded searches for distinguishing databases inform candidate selection.
Sources (1)
- [1]VeriPIE: Checkable Query Plans and Evidence-Guided Selection for Text-to-SQLarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 11:31 PM
We present VeriPIE, a neurosymbolic framework combining language-model generation with symbolic checks.
Financial Artificial Intelligence (AI) governance and model risk management (MRM) emphasize independent validation, transparency, and effective challenge of AI-generated computations.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026Reliable Text-to-SQL in Resource-Constrained Settings with Small LLMs via Iterative Retry Workflows
- Oct 11, 2026Passing Tests, Missing Semantics: Evidence of Interpretive Lock-In in Plan-Based DataFrame Code Generation
- Oct 11, 2026Apertus 1.5 Release Notes
- Oct 10, 2026Counterfactual Regret for Reflective Agents with Lean-Verified Guarantees
- Oct 10, 2026LYNX: Analytical Program Synthesis via Grounding and Closed-Loop Refinement
- Oct 10, 2026vllm-sr/d3