The ultimate AI presentation assistant.

A good slide deck was not enough. An example that came to mind was hard to explain without a picture. Students unfamiliar with Korean had to work harder just to follow the words. I also wanted to help those who missed a concept before the class moved on.

I described those teaching moments to an agent, tried the results and refined them. This is a tool built by the person teaching the class, shaped around the needs I encountered.

“Next slide.” “Show me a photograph of this building.” Spoken requests can reduce interruptions caused by operating the screen. I implemented voice slide navigation in the original presentation tool and added image popups that can be opened and closed by voice.

A photograph can explain an idea faster than a long description. I wanted to show an example that was absent from the prepared slides and then return to the presentation. In the integrated classroom, changes on the instructor screen also reach student screens.

When students spend their attention decoding Korean, they have less attention left for the subject. My spoken explanation becomes live captions, translated into each student’s chosen language. Students can use their native language when supported, or another familiar language.

The six translation languages are English, Simplified Chinese, Japanese, Vietnamese, Burmese and Russian. The Korean original can appear alongside them. Everyone follows the same lesson without having to use the same caption language.

Missing one concept can make the next explanation harder to follow, but stopping the whole class is not always comfortable. Students can ask on their own screens, read an answer and return to the lesson.

The AI assistant answers from the course materials I uploaded and points to the relevant document and page. It uses only materials the student is allowed to access. Students can also send a question directly to me, supporting individual needs within the shared lesson.

Slides contain the main points, but they do not capture every explanation I add during class. I record lecture audio and captions so students can return to those explanations afterward.

Recordings are not published automatically. I can correct captions and exclude unnecessary segments, then review and publish them. Enrolled students can replay the full recording or individual segments while reading the captions.

Lectures include self-corrections and pauses. A literal transcript can be difficult to read, so a local AI model refines Whisper output into readable captions and translations. A glossary tied to the course and slide helps with technical terminology.

If I pause after “The gap between these columns…” and continue with “is wide enough for a person to pass,” the fragments join into one caption. With no continuation, it keeps only what was heard rather than inventing an ending.

A self-correction such as “30—no, 35cm” becomes “35cm.” Checks look for changes to numbers, units and negation, retaining the original when an output fails them. Cleanup and translation can still contain errors, which is why I also review recordings before publication.

The display and audio-input devices can be separate. A phone or computer can capture speech, using a microphone or AirPods when available as an input on that device. My screen focuses on teaching controls; student screens focus on caption language and questions.

I connected local Whisper and language models, separating recognition, cleanup and translation. Audio capture continues while a translation is running, and results appear in spoken order. Relevant terms and concise context help keep processing focused.

The point is who built the tool and why. As the instructor, I identified moments when students needed help and turned those needs into software: presentation controls, translation, questions and review, connected around my class.

My experience running local AI and working with agents helped. I still had to describe the teaching situations, inspect the results and refine them. That process turned one instructor’s idea into a tool that can support different students in the same class.

Voice slide navigation was implemented in the original presentation tool. The current Academy has built-in voice image controls and a slide-control integration point; the features described were developed across different stages. Students can access recordings and captions after instructor review and publication. Actual screenshots use test accounts and synthetic lecture audio. Translation accuracy and latency vary with language, model load and classroom conditions; AirPods and individual devices need classroom validation.