causal-triples

This paper proposes a Damage Cause Encoder for bridge diagnostic agents that uses knowledge triple extraction from manuals and retrieval-augmented fine-tuning with QLoRA, achieving high accuracy with lower memory usage.