In recent years, the rapid transformation of the global economy has been driven by deep technological revolutions. In this context, the concept of new quality productive forces has emerged as a central engine for high-quality socioeconomic development. Nurturing this new vitality requires “new-quality talents” who possess innovative thinking and digital literacy. As an educator and researcher in the field of vehicle engineering, I have observed that higher education must adapt its teaching models to reflect these shifts. In particular, the integration of new quality productive forces with intelligent and digital technologies has become a crucial theoretical and practical issue in educational reform. This article presents a new paradigm for constructing smart courses for electric vehicle design, written from my perspective based on years of teaching reform and practice. The proposed paradigm combines digital resources, AI-assisted platforms, and a SPOC-based hybrid teaching model, which can serve as a valuable reference for instructors who face similar challenges in engineering education.
With the rapid advancement of artificial intelligence (AI), its applications in education have expanded considerably, bringing both opportunities and challenges to instructional reform. AI can deeply analyze existing teaching resources and help instructors systematically plan course content and produce instructional materials. More importantly, AI can generate customized learning paths and resources based on students’ learning status and preferences, thereby promoting autonomous and deep learning. Through the analysis of student learning data, AI also enables accurate assessment of learning outcomes, providing robust evidence for teaching improvements. I believe that the introduction of AI into electric vehicle design courses is not merely an option but a necessity in the post-pandemic and AI-driven era.
Electric vehicle design courses are highly interdisciplinary, involving mechanical engineering, automation, electrical engineering, computer science, and materials science. Such courses face multiple pedagogical challenges, including strong disciplinary intertwining, slow content renewal, monotonous instructional approaches, and one-sided assessment schemes. To solve these problems, I propose an integrated model that combines digital teaching resource construction with a smart course platform, and applies a SPOC (Small Private Online Course) blended teaching mode. This article reports on the design, implementation, and evaluation of this model, with special emphasis on the course “Electric Vehicle Design Fundamentals”. The ultimate goal is to cultivate learners into innovative, digitally competent, and lifelong-learning-oriented engineers who can contribute to the development of new quality productive forces in the automotive sector.
Overall Architecture of the New Paradigm
The overall architecture of the smart course construction paradigm for electric vehicle design under the perspective of new quality productive forces is characterized by four key dimensions: digitalization, intelligence, precision, and personalization. Through digital methods, teaching resources can be enriched, making it possible to push accurate learning content to each student. With intelligent assistance, teacher–student–machine ternary interactions become feasible, while a diversified assessment system helps to improve students’ innovative ability and digital literacy. I designed this architecture with a clear focus on training innovative and constant-learning talents who can thrive in the digital age.
Digital Resource Integration Based on Knowledge Graphs and Modular Design
To meet the graduation requirements and course objectives, I decompose the content of electric vehicle design education into three major modules: basic theoretical modules, design modules, and frontier-technology modules. A large language model is used to deeply mine and analyze existing teaching resources, extracting key terms and performing topic modeling. This process can identify critical information and the relationships between knowledge points. The extracted knowledge points and structured information are then used as nodes in a course knowledge graph. The logical relationships among knowledge points are explicitly defined, forming either a tree or network structure. This ensures that each knowledge point remains sufficiently independent while exhibiting systematic connections. The mathematical representation of a knowledge graph is given by
$$G = (K, E)$$
where \(K = \{k_1, k_2, \dots, k_n\}\) denotes the set of knowledge points and \(E = \{e_{ij}\}_{i,j=1}^n\) denotes edges representing prerequisite, implication, or association relations between \(k_i\) and \(k_j\). For instance, an edge from \(k_i\) to \(k_j\) may indicate that mastering \(k_i\) is necessary before learning \(k_j\). This structured representation allows the curriculum to be dynamically updated by inserting or deleting knowledge points without breaking the entire systematic architecture.
A weighted adjacency matrix can formalize the graph:
$$A_{ij} = \begin{cases}
w_{ij}, & \text{if a meaningful relation exists from } k_i \text{ to } k_j, \\
0, & \text{otherwise}
\end{cases}$$
where \(w_{ij}\) is a semantic weight derived from the strength of relation, such as term-frequency-inverse-document-frequency scores in the original textbook or the frequency of co-occurrence in examination questions. In my practice, I use such weights to distinguish core knowledge points from optional extensions. More generally, the curriculum knowledge graph supports the design of prerequisite chains, scaffolds learning progress, and enables precise placement of digital resources in the online platform.
Based on the knowledge graph, the core knowledge points are identified, including key concepts such as performance indicators of power batteries, operating characteristics of electric motors, and design strategies for energy management systems. I record short videos for each core knowledge point so that students can easily preview and review. Every module contains a set of videos, including introductory explanations, animations of working principles, and typical component structural analyses.
AI-Assisted Smart Teaching Platform Construction
Using the constructed knowledge graph as the foundation, I organize the core knowledge point matrix and record short videos for those points. These resources are placed on an online education platform built on an AI-enhanced system that I refer to as the “AI + smart course platform”. The platform supports self-learning before class and after class. Differing from conventional learning management systems, this platform is equipped with AI that can generate a diversified question bank automatically based on the knowledge graph. This question bank is used both before and after class to assess students’ mastery of the core knowledge points identified in the graph. The AI engine can also augment project-based teaching by suggesting personalized project topics with appropriate degrees of challenge, thereby satisfying students’ diverse interests and fostering well-rounded development.
AI + SPOC Personalized Teaching Practice
Based on the SPOC model, I combine the advantages of offline classroom teaching and online course learning. The entire teaching flow is designed as “online import – online/offline participation – online assessment and feedback”. AI technologies allow pre-class optional tasks, in-class online quizzes, and automatic recording of learning behaviours. By using these data, teachers and students can obtain real-time information about knowledge mastery and receive personalized and accurate recommendations for exploratory project topics. In this context, AI is used to support teacher–student interaction and feedback, forming a teacher–machine–student ternary interaction scene. Such a scene is highly effective for collecting multi-dimensional information about each learner’s status. A high-level expression of this online, offline, and assessment loop can be written as
<$$>
M_{\text{SPOC}} = \Big\langle I_{\text{online}}^{(t)},\; P_{\text{hybrid}}^{(t)},\; D_{\text{assessment}}^{(t)} \Big\rangle_{t=1}^{T}
$$
where \(t\) indexes a teaching period, \(I_{\text{online}}^{(t)}\) represents the online import activities in period \(t\), \(P_{\text{hybrid}}^{(t)}\) denotes hybrid participatory activities, and \(D_{\text{assessment}}^{(t)}\) represents the assessment results. The recursive relationship \(M_{\text{SPOC}}^{(t-1)} \rightarrow M_{\text{SPOC}}^{(t)}\) allows the teaching system to adapt itself when a new period begins.
Diversified Process Assessment System for New-Quality Talents
Traditional examinations that only measure memorized knowledge cannot fully evaluate the capabilities needed in a new-quality productive force era. Therefore, I design a comprehensive assessment framework with AI-supported standards. The standards include online course participation, collaborative attitude in group projects, problem-solving skills, technological application capabilities, and so on. The evaluation is built on multiple channels, including learning-behavior records, project reports, homework, final closed-book examinations, online quizzes, optional tasks, and group discussions. Each channel contributes to the final score in a weighted manner. This approach ensures more accurate and complete evaluation outcomes and creates positive washback on students’ continuous learning.
Course Construction and Teaching Reform Design: A Case Study of “Electric Vehicle Design Fundamentals”
Digital Teaching Resource Construction
In this section, I take the course “Electric Vehicle Design Fundamentals” as a concrete example to illustrate the construction process. According to the professional training program, the course objectives must state clearly what students should be able to do after completing the course. Using a large language model, I feed the course syllabus, digital textbooks, previous examination papers, and other teaching resources into an AI-assisted text-mining system. The system identifies key information and builds the core module architecture. Combined with development trends in the electric vehicle industry, the course content is decomposed into four teaching modules: battery, motor, electronic control, and vehicle integration. These modules are related through the intrinsic connections among their contents. The final result is a networked knowledge graph that covers all vital concepts, as summarized in Table 1.
| Module | Core content area | Representative knowledge points |
|---|---|---|
| Battery | Power battery systems | Battery capacity, state of charge, thermal management, equalization circuits, solid-state battery, LFP blade battery |
| Motor | Drive motors and actuators | Permanent-magnet synchronous motor, induction motor, reluctance motor, motor torque-speed characteristics, loss analysis |
| Electronic control | Controllers, power electronics, energy management | Vehicle control unit, motor controller, DC-DC converter, energy recovery strategy, energy management strategy |
| Vehicle integration | Vehicle-level design and dynamic performance | Powertrain parameter matching, hybrid architecture design, pure electric architecture, fuel cell architecture, vehicle lightweight design |
The knowledge graph for the entire course can be expressed as
$$K_{\text{EV}} = \{ k_{\text{battery}}, k_{\text{motor}}, k_{\text{electronicControl}}, k_{\text{vehicleIntegration}} \},$$
where each module \(k_{\text{battery}}\) itself is a subgraph \(G_{\text{battery}} = (K_{\text{battery}}, E_{\text{battery}})\). For example,
$$K_{\text{battery}} = \{ \text{capacity}, \text{voltage}, \text{thermal behavior}, \text{aging}, \text{safety}, \text{battery management system} \}$$
and \(E_{\text{battery}}\) contains edges such as “aging leads to capacity fading” and “thermal limits determine battery pack design.” This hierarchical representation is both scalable and easy to implement in an online platform.
For each knowledge point, I assign a difficulty level \(d \in \{L_1, L_2, L_3\}\) and a transparency level indicating whether the point should be seen by all students or only as an optional advanced topic. The combined learning objective for module \(m\) can be defined as
$$O_m = \sum_{i=1}^{n_m} \lambda_i \, c_i,$$
where \(c_i\) is the mastered level for the \(i\)-th knowledge point of module \(m\), and \(\lambda_i\) stands for the curriculum weight. In my assessment, \(\lambda_i\) is initially derived from the credit hours assigned to each module and is further refined according to industry demands.
Construction of a Hybrid Smart Course Platform
The “AI + smart course platform” used in this case is an online education platform that enables the instructor to upload lecture notes, knowledge graphs, required videos of core knowledge points, optional videos of frontier technologies, extended references, exercise banks, and project resources. The platform also supports various teaching-management functions, such as releasing course tasks (including both required and optional tasks), launching discussion forums, guiding students to select design topics individually, and carrying out chapter tests. Students interact with these resources by watching videos, participating in online discussions, submitting assignments and tests, and receiving real-time feedback. From the platform, I can inspect records such as video progress, question-answer frequency, time spent on each document, and quiz results. Based on these behavioural data, I can promptly understand students’ interests and learning progress. For example, the overall learning process is governed by
$$Q_{\text{effective}}(t) = \frac{1}{t}\sum_{\tau=1}^{t} \sum_{i \in I_{\tau}} v_i(\tau) \cdot r_i(\tau)$$
where \(v_i(\tau)\) is the volume of learning activity on resource \(i\) in time interval \(\tau\), and \(r_i(\tau)\) is the percentage of students who accessed that resource at least once. The larger the value of \(Q_{\text{effective}}\), the more strongly the resources capture the attention of learners.
One essential feature is the personalized learning path. For each student, the platform calculates a recommendation score for every knowledge point based on their historical performance:
$$R(u, k) = \alpha \cdot \mathrm{preference}(u, k) + \beta \cdot \mathrm{difficulty}(k)^{-1} + \gamma \cdot \mathrm{relatedness}(k, k_{\text{last}})$$
where \(\alpha\), \(\beta\), and \(\gamma\) are non-negative weights that sum to unity. The preference score is acquired from clicks, bookmarks, and voluntary quizzes; the difficulty term balances challenge and comprehensibility; and the relatedness term ensures continuity from previously learned knowledge point \(k_{\text{last}}\). For instance, if a student performs poorly on thermal management, the algorithm may recommend first reviewing heat-transfer fundamentals before assigning a project on battery cooling system design.
Intelligent Teacher–Machine–Student Ternary Teaching Design
In the case study, I adopt an improved BOPPPS model, where BOPPPS stands for Bridge-in, Objective, Pre-assessment, Participatory Learning, Post-assessment, and Summary. The SPOC blended teaching process for “Electric Vehicle Design Fundamentals” is divided into three broad phases: importing, interactive teaching, and assessment. Table 2 summarizes the distinctive roles of teachers and students in each phase, and highlights the AI platform as the machine partner.
| Phase | Teacher role | Machine (AI) role | Student role |
|---|---|---|---|
| Online import | Produce digital resources, post pre-course tasks, answer questions | Collect resource-viewing data, recommend prerequisite knowledge, assign individualized pre-class exercises | Study knowledge graph, complete required tasks, finish optional frontier tasks |
| In-class/online participatory learning | Deliver lectures, foster discussions, guide project groups | Run real-time polls, random quizzes, display learning analytics dashboards | Answer questions, join group discussions, submit survey responses |
| Post-assessment and feedback | Design chapter tests, allocate homework, issue project tasks | Auto-grade quizzes, generate personalized feedback, record mistakes | Do online homework, take timed quizzes, write project reports |
The BOPPPS module can be represented as a sequence of teaching stages
$$S = [B, O, P_{pre}, P_{part}, P_{post}, S_{sum}]$$
where each stage is implemented online, offline, or in a blended form. In my design, the pre-assessment stage \(P_{pre}\) uses AI to generate a diagnostic quiz that checks if students possess necessary prior knowledge. The participatory learning stage \(P_{part}\) combines teacher explanations, group discussion, and machine-generated mini-case analyses. The post-assessment stage \(P_{post}\) also involves a machine-graded quiz and a project task. Because AI can generate quizzes with different difficulty levels for different groups, I can realize adaptive teaching more easily than with traditional methods.
One concrete example is the process of personalized project topic selection. Students are expected to select a complete design exercise topic for an electric vehicle subsystem. Machine analysis of students’ earlier learning behaviour provides data-driven topic recommendations. Table 3 lists example project themes in four groups and the corresponding major focus.
| Area | Project topic | Typical emphasis |
|---|---|---|
| Battery pack | Lightweight battery pack and structural design | CAD/CAE, material selection, crash safety |
| Thermal management | Electric vehicle battery thermal management system design | Heat-transfer simulation, cooling circuit design, control strategy |
| Battery management | Battery cell equalization method and control strategy | Power electronics, embedded coding, algorithm design |
| State estimation | Battery health state prediction using machine learning | Data analytics, regression models, model verification |
| Energy management | Energy management system for a hybrid electric vehicle | Optimal control, dynamic programming, rule-based logic |
| Vehicle controller | VCU design for an electric vehicle | System architecture, CAN bus, functional safety |
| Regenerative braking | Energy recovery system for urban electric vehicles | Braking force distribution, motor/generator mode, efficiency map |
| Power converter | DC–DC converter efficiency optimization for EVs | Circuit simulation, MOSFET/IGBT selection, thermal analysis |
| Hybrid system | Parameter design of a parallel hybrid powertrain | Simulation of vehicle dynamics, fuel economy calculation |
| Full-electric system | Powertrain parameter design for a pure electric bus | Driving cycle analysis, motor sizing, battery sizing |
| Range-extended EV | Range-extended electric vehicle powertrain matching | Range calculation, APU control, energy consumption |
| Fuel-cell vehicle | Fuel cell vehicle powertrain design | Fuel cell stack model, hydrogen storage system, hybrid battery sizing |
| Motor control | Parameter optimization for a permanent-magnet synchronous motor | Finite-element magnetic simulation, field-oriented control |
| Induction motor | Rotor resistance identification of an induction motor | System identification, sensorless speed observation |
| Reluctance motor | Magnetic reluctance torque optimization in synchronous reluctance motors | Rotor geometry, torque ripple, finite-element analysis |
| Drive control | Electric vehicle drive motor control strategy | PI controller tuning, torque-speed envelope, hardware-in-the-loop testing |
When students interact with the system, the personalized recommendation function can be expressed by the following probability model:
$$\Pr(\text{student } u \text{ selects topic } p) = \frac{\exp(\theta_u^{\top} \phi_p)}{\sum_{p’ \in \mathcal{P}} \exp(\theta_u^{\top} \phi_{p’})}$$
where \(\theta_u\) is a \(d\)-dimensional vector describing the student’s proficiency profile, \(\phi_p\) is the feature vector representing the competency demands of topic \(p\), and \(\mathcal{P}\) is the set of available topics. This softmax formulation formalizes how AI matches students with challenging but feasible design tasks, instead of assigning one uniform topic to every student.
Multi-Source and Process-Oriented Assessment System
For a new-quality-talent cultivation environment, process assessment should not rely solely on the final exam. In this course, the overall score is divided into an online part and an offline part, each taking 50%. The calculation equation is
$$T_{\text{total}} = 0.5\,O_{\text{online}} + 0.5\,O_{\text{offline}}.$$
The online evaluation \(O_{\text{online}}\) further contains five components: after-class homework, class participation, chapter quizzes, personalized project report, and pre-class tasks. In the actual implementation, I use
$$O_{\text{online}} = 0.05\,C_{\text{pre}} + 0.05\,H_{\text{hw}} + 0.05\,P_{\text{class}} + 0.10\,Q_{\text{quiz}} + 0.25\,R_{\text{project}},$$
where \(C_{\text{pre}}\) is the pre-course task score (including both optional and required assignments), \(H_{\text{hw}}\) is the regular homework score, \(P_{\text{class}}\) measures classroom participation, \(Q_{\text{quiz}}\) represents chapter tests, and \(R_{\text{project}}\) corresponds to the individual project report. The offline examination \(O_{\text{offline}}\) is a closed-book final examination that consists of conceptual questions and design-oriented problems. To calculate the standardized project score, I further define
$$R_{\text{project}} = 0.3\,R_{\text{proposal}} + 0.4\,R_{\text{final}} + 0.3\,R_{\text{presentation}},$$
where \(R_{\text{proposal}}\) evaluates the feasibility of the selected topic, \(R_{\text{final}}\) evaluates the depth and correctness of the engineering solution, and \(R_{\text{presentation}}\) is obtained by peer and teacher rating of the final defence. The weights were determined after two teaching cycles and reflect the pedagogic emphasis on system design, digital simulation, and communication skills.
Additionally, I record the AI-generated learner profile
$$P_{\text{learner}}(u) = \Big( \bar{t}_{\text{video}}, \bar{t}_{\text{active}}, q_{\text{forum}}, e_{\text{error}}^{(1)}, e_{\text{error}}^{(2)}, \dots, e_{\text{error}}^{(m)} \Big)$$
for every student \(u\). Here, \(\bar{t}_{\text{video}}\) is the average video-completion time, \(\bar{t}_{\text{active}}\) is the average active learning time, \(q_{\text{forum}}\) is the number of forum posts, and \(e_{\text{error}}^{(r)}\) is the count of repeated errors on the \(r\)-th error-prone knowledge point. This profile feeds back into the recommendation model of the platform, enabling a closed evaluation loop.
Implementation Effects and Continuous Improvement Measures
After one full teaching cycle, I collected learning-behaviour data from the smart-course platform for all registered students. The data allowed me to identify popular optional learning resources, common misconceptions, and the effects of different assignments on learning outcomes. For example, Table 4 displays the statistics of the online-learning hotspot in the optional frontier module. For each trend topic, the table records the click-through rate and the number of AI-facilitated Q&A exchanges.
| Hot topics | Click-through rate (%) | Number of Q&A interactions |
|---|---|---|
| Solid-state batteries | 95.3 | 74 |
| Blade batteries | 94.5 | 62 |
| Hairpin winding motors | 91.2 | 87 |
| Fuel cell technology | 88.6 | 53 |
| Regenerative braking systems | 87.9 | 49 |
From Table 4, I notice that students have a strong preference for frontier topics such as solid-state batteries and blade batteries. This signals that the optional module should be continuously updated to include the newest industrial developments. In the next cycle, I plan to add videos on cell-to-pack technologies and high-rate fast-charging chargers. Furthermore, I will design more project topics related to these hot areas so that students can connect their engineering designs with the newest evolutions in the electric vehicle industry.
Meanwhile, the chapter-quiz statistics presented in Table 5 reveal several error-prone knowledge points. The score rate is the percentage of students who correctly answered all questions for that knowledge point. The relatively low score rate on the calculation of power batteries highlights a challenge in deriving battery energy and power requirements. As a remedy, I can decompose the relevant knowledge videos into smaller segments. For instance, “battery-related calculations” can be divided into capacity calculation, energy density calculation, C-rate analysis, and thermal-loss calculation. Each sub-video is accompanied by a mini- exercise. The AI platform can be trained to recognize typical mistakes in student input and offer polished supportive explanations. This helps students quickly locate their own misunderstandings.
| Knowledge point | Score rate (%) | Improvement strategy |
|---|---|---|
| Power-battery related calculation | 80.0 | Produce smaller video units, add step-by-step calculation templates, use AI assistant for targeted exercises |
| Motor parameter calculation | 82.0 | Offer interactive motor-sizing spreadsheets, generate adaptive electric motor design examples |
| Hybrid electric vehicle architecture drawing | 83.0 | Add more simulation tools for different architectures, increase in-class discussion, compare series-parallel configurations |
| Powertrain parameter matching | 83.0 | Design guided design projects, use real electric vehicle data, create matching exercise banks |
| Energy management strategy | 85.0 | Introduce rule-based and optimization-based strategies via case studies, let students simulate driving cycles |
For the knowledge point “hybrid electric vehicle architecture drawing”, 17% of students could not accurately draw the architecture. To address this, I will expand the in-class discussion section to include demonstrations of various hybrid topologies, and I will push customized review materials to weak students. The expected outcome is to raise the score rate from 83% to at least 90% in the next implementation. This kind of data-driven continuous improvement is a distinctive advantage of smart-course platforms.
In addition to quantitative quizzes, the platform records qualitative information from group discussions and AI-mediated reflection exercises. I define the normalized improvement index for each error-prone point \(j\) as
$$\mathrm{Improvement}_j = \frac{ (\mathrm{ScoreRate}^{(t+1)}(j) – \mathrm{ScoreRate}^{(t)}(j)) }{ \max\{ \mathrm{ScoreRate}^{(t)}(j), 1 – \mathrm{ScoreRate}^{(t)}(j)\} } \times 100\%,$$
where \(t\) denotes teaching cycle. This normalization ensures that improvement is measured fairly even for knowledge points that started at high levels. In my future evaluation, I will apply this formula to each of the five error-prone points listed in Table 5 and compare the results across two cycles.
Another important metric is the diversity of resource access. The platform calculates a Gini-like coefficient of resource distribution. Define \(p_i\) as the proportion of total access received by resource \(i\) over one teaching cycle. The concentration index is
$$C_{\text{resource}} = 1 – \sum_{i=1}^{N} p_i^2.$$
A high value of \(C_{\text{resource}}\) indicates that students access a wide variety of videos, quizzes, and articles, which demonstrates broad engagement rather than narrow repetition. In my course data, the resource concentration index increased from 0.84 in the previous traditional e-learning environment to 0.91 in the new AI-assisted platform, showing that personalized recommendations encourage exploration of diverse electric vehicle subjects.
Beyond that, I use an entropy-based learning interest measure. If a student accesses a set of topics with probabilities \(p_{1}, p_{2}, \dots, p_{n}\), then the learning-path entropy is
$$
H_{\text{learning}}(u) = -\sum_{i=1}^{n} p_i \log_2 p_i.
$$
A higher entropy means that the student balances their attention between several relevant areas rather than staying in one narrow niche. At the beginning of the course, the mean entropy was 2.1 bits, while at the end it reached 3.6 bits. This shift indicates that the knowledge-graph navigation and AI recommendations helped students develop a broader understanding of electric vehicle design.
Through the above analyses, I observe that the proposed model transforms the conventional teacher-directed mode into a data-rich, learner-centred environment. The construction of course knowledge graphs and the smart-course platform allow conventional instruction in electric vehicle design to shift toward digitalization and intelligence. AI-assisted instructional design encourages teachers to keep learning new digital technologies and to improve their professional and pedagogical proficiency. It simultaneously enhances the efficiency of classroom teaching and stimulates students’ creative thinking and problem-solving skills. From my own teaching reflections, I find that the new paradigm enables students to perceive the frontier direction of electric vehicle engineering, encourages them to ask questions, and gives them the confidence to propose original design solutions. Moreover, the explicit knowledge-graph structure allows them to see where they are in the large landscape of electric vehicle engineering, preventing the common feeling of being lost in a fragmented course.
Conclusion
In summary, I have developed and tested a new paradigm for smart-course construction in electric vehicle design from the perspective of new quality productive forces. The paradigm incorporates digital teaching-resource construction, an AI-based smart-course platform, and efficient teacher–machine–student ternary teaching design. A key contribution of the work is the use of a large language model to transform scattered textbooks and academic resources into a structured knowledge graph. The graph not only supports modular teaching in electric vehicle design but also enables personalized learning paths for students with different backgrounds. The AI platform uses this graph to recommend specific resources and project topics, while simultaneously recording learning behaviour for continuous feedback. The process assessment system combines online quizzes, homework, project reports, class participation, and final examinations in a diversified manner. The case study of “Electric Vehicle Design Fundamentals” shows that students are highly interested in frontier technologies, such as solid-state batteries, blade batteries, and hairpin winding motors. The data-driven analysis of error-prone knowledge points has directly guided subsequent teaching improvements, such as splitting videos into smaller pieces and training the AI assistant to answer common calculation mistakes. In the future, I plan to add more advanced natural language processing tools to support semiautomatic question generation for electric vehicle design topics. I also intend to integrate cross-university collaborations so that students from different disciplines can work on realistic electric vehicle design projects through the same platform. Such an expansion will help to cultivate talents who possess both specialized engineering knowledge and flexible digital competence, thereby meeting the far-reaching requirements of new quality productive forces.
The adoption of this paradigm is not limited to one course. I believe that the same architecture of knowledge-graph-based smart-course construction can be transferred to other engineering disciplines, such as robotics, intelligent vehicles and energy systems. With the continuous evolution of artificial intelligence and new quality productive forces, higher education must embrace a lifelong-learning ecosystem that respects individual differences and provides timely intelligent feedback. The result is an agile and human-centred teaching environment where students, teachers and AI work together to achieve high-quality engineering education. My experience with the electric vehicle course demonstrates that this vision is not only plausible but also effective. The evidence comes from improved learning hotspots, higher engagement indexes, correctable error patterns, and a positive change in the learning-path entropy. Admittedly, further research is needed to relate these quantitative indicators to high-level competences such as creativity and engineering judgment. Still, the promising trajectory of the smart course encourages me to continue refining every component of the paradigm and to contribute to the larger movement of digitally transformed engineering education.