Semantic Relational Types of SQL Queries and Applications to AI Agent Tool Selection
Bibliographic record
Abstract
We introduce Semantic Relational Types (SRT), an extension to the relational type system that enriches traditional SQL data types with semantic annotations. SRT provides LLMfriendly type annotations for relational queries expressed in SQL, enabling effective integration between database systems and large language models. Our primary contribution is an algorithm for automatic SRT generation through type inference of arbitrary SQL queries. The algorithm performs recursive decomposition of SQL statements into nested subqueries and clauses, utilizing LLMs with structured prompt engineering to generate semantic descriptions. We also introduce semantic type checking, which leverages semantic similarity measures to verify pairwise compatibility between SRTs, extending beyond traditional structural compatibility. We focus on applying SRT to LLM tool-calling scenarios where tools must be selected based on natural language queries (NLQ). In our framework, tools are implemented as parameterized SQL queries, and their SRTs in JSON schema format serve as tool descriptions provided to the LLM. This enables the LLM to perform semantic type checking and select the most compatible tool for any given NLQ. Our experimental evaluation demonstrates that SRT-based tool descriptions significantly improve tool selection accuracy compared to raw SQL queries. Tool-set refinement using semantic type checking between NLQs and tool SRTs yields superior results compared to inferred SRTs alone. The approach is validated using TPC-DS with 99 tools and multiple NLQ styles.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.011 | 0.037 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.005 | 0.006 |
| Science and technology studies | 0.001 | 0.003 |
| Scholarly communication | 0.007 | 0.011 |
| Open science | 0.004 | 0.005 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".