Research questionHow can natural-language interfaces generate reliable SQL and GraphQL without runtime LLM inference?Natural-language database interfaces commonly target only SQL and rely on an LLM for every query, yet syntactically valid outputs may still express the wrong meaning. Supporting multiple query languages while keeping inference lightweight also requires a way to signal possible semantic errors at runtime.