AIriskEval-edu Demo: Auditing Pedagogical Risks in Instructional Explanations

AIriskEval-edu Demo audits pedagogical risks in instructional explanations, and its self-hosted Llama 3.1 8B evaluator outperforms GPT-5.5 on most reported metrics. The platform gives judgments, confidence, and rationales across five dimensions, including factual accuracy, student age-appropriateness, and ideological bias. The evaluator is fine-tuned on the K-12 dataset AIriskEval-edu, allowing educational institutions to audit within their own infrastructure.