TG-RAG: A Retrieval-Augmented Framework for Reasoning Guidance in Specialized Domains

Enhancing Large Reasoning Models (LRMs) for specialized domains remains a critical challenge. While recent industrial frameworks attempt to encapsulate Standard Operating Procedures into modular "skills" for dynamic retrieval, utilizing them via context engineering often proves insufficient for complex workflows, leading to "Cognitive Drift." To mitigate this, we propose $\textbf{Thought Guidance-Retrieval Augmented Generation (TG-RAG)}$, a Retrieval-Augmented framework that effectively steers the generation process without relying solely on the model's self-correction. Built upon an Expert Procedure Graph (EPG) that formalizes unstructured SOPs, the framework uniquely employs a dynamic $\textbf{``Interrupt-Retrieve-Generate" (IRG)}$ mechanism to actively inject step-specific directives into the model's reasoning process. Extensive evaluations show that TG-RAG achieves competitive performance, demonstrating advantages in specialized domains by ensuring faithful adherence to domain SOPs. Code is available at https://github.com/V1ncent-S/Thought-Guidance.

When we ask artificial intelligence to handle highly specialized tasks—like diagnosing a disease or analyzing financial fraud—it can easily lose its train of thought and deviate from strict professional rules. We call this "cognitive drift." Current solutions try to fix this by giving the AI more background reading upfront, but this doesn't guarantee the AI will actually follow the correct logical procedure. To solve this, we developed a system called TG-RAG. Instead of just handing the AI a thick manual to read beforehand, our framework acts like an active, step-by-step supervisor. We map out expert workflows as logical flowcharts. Then, as the AI generates an answer, our system physically interrupts it at every critical decision point to inject the exact rule and necessary data for that specific step. This step-by-step guidance prevents the AI from wandering off-track. By transforming passive suggestions into active, enforceable rules, our method significantly boosts AI accuracy, making it much safer and more reliable to deploy in high-stakes industries where strictly following procedures is critical.