This research introduces a novel approach for precisely controlling Large Language Models during the inference phase to ensure reliable requirement verification in critical engineering applications.

• Enables fine-grained control of LLM outputs without retraining
• Demonstrates application with Capella SysML models for space mission validation
• Achieves higher reliability than conventional prompting or fine-tuning
• Provides a framework for dynamic adjustments to meet engineering precision requirements

For aerospace organizations, this technique represents a significant advancement in integrating AI into safety-critical systems engineering workflows, where accuracy and verification are paramount.

Inference-Time Intervention in Large Language Models for Reliable Requirement Verification