AI-Driven Intelligent Assessment for Higher Education: From Data Engineering to Trustworthy Online Evaluation

This article focuses on the implementation of "intelligent evaluation"in higher education teaching reform. Based on a computer systems perspective, it constructs an integrated data-model-feedback implementation path. Initially, it builds a teaching behavior data pipeline, connecting LMS/exam and interaction logs to achieve a complete process covering collection, cleaning, desensitization, feature engineering, and feature repository management. Secondly, at the model level, it achieves multi-model collaboration through sequence modeling (Transformer/TCN) and graph modeling (course knowledge graph/peer learning graph). It also utilizes calibration (temperature scaling/ECE) and uncertainty assessment (MC-Dropout/Deep Learning). Ensembles maintains the reliability of teaching decisions; it again relies on streaming inference and feature snapshots to achieve minute-level online evaluation and personalized feedback, and uses A/B and quasi-experimental designs to assess teaching effectiveness; in terms of systems engineering, it relies on containerization and service mesh to achieve scalable deployment, and uses feature drift/concept drift monitoring and small-step rollback to ensure model governance and traceable auditing. Compared with baseline manual and offline practices, this solution has achieved significant optimization in key indicators such as homework grading latency, learning effectiveness improvement rate, feedback timeliness, and teacher workload, and also highlights the reusability of end-to-end MLOps in education scenarios.