Tag
PatchBoard replaces natural-language dialogue in LLM multi-agent systems with validated JSON Patch mutations over a shared structured state, achieving higher success rates and significantly lower token usage on ALFWorld benchmarks.
This paper presents a schema-grounded natural language interface for transportation safety analysis that uses a large language model to interpret user queries while preserving deterministic execution against an authoritative database. The framework is evaluated on a Massachusetts transportation safety database, successfully executing all queries and correcting errors in 29% of cases, demonstrating a practical approach to broadening access to safety data.