A university teaching system becomes an instructor’s own platform

Lecture materials were prepared as files, and quizzes were operated on a separate server. Assignments and school attendance also had to be checked on their respective screens. Between conducting the class, tasks such as transferring materials, comparing results, and re-entering the same information were interspersed.

The same was true for students. When the places for reading materials, solving problems, and submitting assignments are separated, unnecessary effort is required to follow the flow of the class. For foreign students listening to explanations in Korean, a tool to immediately verify missed sentences was also necessary.

What I wanted was clear.An instructor-focused space where instructors could concentrate on explanations and feedback, and students could view materials, ask questions, and solve problems within the same class.I described these requirements to GPT Astra and created UPMI Academy.

The video below shows the actual Academy app running in a separate demo environment. The slides clearly change from a blocked ground floor to an open space with columns, and then to a floor plan of pedestrian routes, with thick blue underlines drawn on both screens simultaneously. As the instructor turns the slides and draws underlines, the student screen changes along with it. Students check the English subtitles at the bottom and ask the AI teaching assistant about 'piloti' if they do not understand.

What I wanted from an electronic whiteboard was not just a large screen, but features that make teaching easier: writing on lecture materials, highlighting important parts, viewing the same screen as students, and leaving explanations after class.

So I created a web-based electronic whiteboard. It includes pen and highlighter tools, color and thickness adjustment, stroke eraser, undo, multiple whiteboards, and slide saving. Each slide retains its whiteboard content, allowing you to review what was written when returning to that slide.

You can use an iPad with Apple Pencil for writing and AirPods-like microphones recognized by the browser for voice explanations.Students view the same slides and whiteboard content on their laptops or tablets. In the classroom, you can simply use existing monitors or projectors. Handwriting feel and microphone input vary by device/browser, so check them before class.

I also connected features to change slides via voice or load related building photos. If short commands like 'next' are not recognized well, I adjusted the recognition start point and command processing order. My lecture habits became the improvement standard.

When listening to Korean explanations, it is difficult for all students to immediately raise their hands when they encounter unfamiliar concepts. If you can re-read what was just heard via subtitles, briefly ask the AI teaching assistant, and then return to the class, an additional method of keeping up becomes available.

The scope of answers was also important. Rather than explaining at length based on any material,I created it so that it answers based on the course materials I selected and states that there is no answer if there is no basis. The summary after class also distinguishes between actual spoken content and supplementary explanations during review. Currently, summaries are in an instructor review/export format and are not automatically distributed to students.

If, after students submit their assignments, I have to go back and forth between other folders and tables to organize grades, having only an upload function is not sufficient. In the academy, we prepared assignments and evaluation criteria, connected them to confirm submitted materials, and enabled the input of grades and feedback.

Students submit files and check their submission status. I view submissions and evaluation criteria in the same flow and evaluate immediately. Draft evaluations are temporarily saved, and after publication, students can view their own evaluations. Submission records are also kept to allow verification of changes.

Here, 'immediate evaluation' means that the instructor can review and provide feedback right away.I did not state that AI automatically grades assignments. I distinguished between evaluating assignments that require judgment and grading quizzes that can be processed automatically based on answer rules.

To avoid having students download files and switch apps every time they want to read materials, I provided an e-book reading screen within the homepage where pages can be turned. Students can open materials appropriate for their enrollment permissions and recheck necessary chapter tables.

The student screen displays watermarks using personal identification information and dates, restricting access to the original file and limiting copying and printing. Instructors view materials without student watermarks so they can focus on lectures.

In the example below, you can identify 'Kim Ha-neul (Demo)', the identifier value, and the date. Watermarks are a mechanism to prevent unauthorized copying and redistribution and to verify the source. I do not describe them as a feature that completely blocks operating system screen captures or camera photography.

I did not start development having designed all features from the beginning. I reviewed existing materials and quiz systems, first clarifying what I do in class and what students need to do. GPT Astra assisted in translating those requirements into screens, permissions, data connections, and operational rules.

I did not stop after creating just the feature to upload materials. I actually read them, took notes, turned chapter tables, and submitted them via the exam account to make corrections. I also verified whether students could not change the instructor's notes, whether evaluations from other students were hidden, and whether records could be retrieved again after class.

As small operational conditions accumulated—such as short voice commands, saving blackboard notes, line breaks in foreign language subtitles, and the timing of releasing evaluations, similar to attendance weeks after a cancelled class—it became a system tailored to my teaching.

The change I felt most significantly was not the number of features. It is the ability to explain the thought 'It would be better to teach like this,' write it out directly, and then revise it. The class is designed by me and created with students. AI helps with the actual work of creating the tools needed for that class.