Online–offline blended learning has become a mainstream direction of higher-education reform

yet in courses that demand strong logical reasoning and programming skills—such as Spark Big Data Architecture—existing smart teaching tools still suffer from shallow use of learning data

weak real-time and personalized feedback

imprecise assessment of learning states

and a heavy analytical burden on teachers. To address these problems

this study designs and implements a blended teaching model supported by an AI teaching assistant (AI TA) and a course knowledge graph. Guided by course objectives

the model couples online autonomous learning with offline classroom instruction under a "teacher-led

student-centered

AI-assisted" framework. A Spark knowledge graph structures fragmented resources into a coherent knowledge network

while the AI TA—built on a large language model and a curated course knowledge base—provides round-the-clock question answering

source-grounded explanations

resource recommendation

and personalized support across the pre-class

in-class

and post-class stages. The model was applied in an undergraduate Spark course and evaluated through questionnaires and semi-structured interviews. Results show high student satisfaction: 92% strongly agreed with the overall model

100% found it superior to traditional teaching for mastering content

and all respondents were willing to keep using it. The findings indicate that pairing a knowledge graph with an AI TA can effectively lower the comprehension threshold for abstract concepts

improve practice efficiency

and strengthen engagement

while the AI TA is best positioned as a support tool rather than a replacement for the teacher.