International Journal of Scientific Research and Engineering Development

The performance of Large Language Models (LLMs) on Open Domain Question Answering (ODQA) datasets is strong in generating human-like and contextually relevant responses. Nevertheless, these models are vulnerable to reliability problems like hallucinations, logical inconsistencies, arithmetic errors, and lack of confidence transparency. These limitations are especially critical in STEM and academic settings, where factual understanding, logical reasoning, and explainability are essential. To address these challenges, this paper introduces ICE-GRAPH (Integration of Code-Guided Engine with Graph-Reasoned Architecture Pipeline), a neuro-symbolic architecture, which consists of RetrievalAugmented Generation (RAG) as a contextual grounding method, code-guided reasoning as a structured computational and procedural validation framework, and multi-hop knowledge graph reasoning as a conceptual relationship validation system between academic entities. This integrated approach ensures both procedural correctness as well as semantic consistency of generated responses. Moreover, symbolic validation modules ensure that the arithmetic accuracy, logical consistency and unit correctness of STEMrelated responses are verified. A perplexity-based confidence calibration mechanism is employed to estimate the reliability of generated outputs. Low-confidence or unstable responses initiate an auxiliary regeneration plan involving correcting responses with new retrieval and reasoning limitations. Large-scale experimental assessment shows large improvements over baseline models. The ICE-GRAPH model decreases hallucination errors by 31.5 percent, achieves query classification accuracy of 91 percent, answer accuracy of 92.7 percent, with a trustworthiness score of 90.3. These findings highlight the effectiveness of combining code-guided reasoning with knowledge graph validation to produce reliable AI-based educational assistants and intelligent tutoring systems.

Akshaya R, Ashwatha D, Aishwarya E V, Vanitha A,"Integration of Code-Guided Engine with Graph-Reasoned Architecture for Hallucination Detection in Student Question Answering Systems" International Journal of Scientific Research and Engineering Development, V9(2): Page(1997-2006) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.