Thank you. I'm Baptiste. I'm co-founder of Loginworks with Thiago and Philippe. I have Thiago here helping me with a technical question for the end of this talk and the Q&A.
Well, thanks for being present today and thanks. We are grateful for the invites of MindStone. Here, we're going to present you a project that is actually still under development. The last delivery was actually today on our QA environment, so let's hope that the demo or I won't fail. But this is it.
We're going to try to present you, if I see my mouse, well, how to scale learning with AI-powered course creation. So to tell you shortly about us, Loginworks, we are an AI boutique based in Lisbon.
Well, the idea is that we see that a lot of AI, and sorry in advance, I'm gonna say AI, AI a lot in this presentation, but that's part of it, I guess. 1And we see that there is a lot of AI that's used in company, but mostly at the collaborator level.
So maybe using ChatGPT, using Copilot on an individual level. And when we don't have a skill up to help leveling up the skills of people to understand how they can use it, they maybe need some more guidance to it.
And what we try to do is to integrate AI at software level or at business process level to enhance the work of, well, different jobs. Well, we have very different background and actually I'm the non-tech here.
Our client is a US-based e-learning company which is doing B2B courses mostly. They also do B2C.
So what's important is to know that they have their own platform. It's an e-learning company.
They are course creators, so they have instructional designers, SMEs, subject matter experts to create the course. They do the course and put the course online, but they also export it for other platforms like Coursera or Udemy, for example.
1The project here was to integrate AI in the company. The good thing was that this company already had their roadmap and their idea of what they wanted to create.
Two main topics were the content creation, so AI-assisted assessment creation and AI-assisted course creation. And the other one, it's an AI assistant for teacher and students.
So here we work with them since June, actually, and we had two big work packages. One first work package was to develop an API-based AI agent to create formative and summative assessments based on relevant inputs. The idea is to allow a generation of a batch of an assessment.
And the work package tool was to go deeper in the creation of the course with AI generation of the course outline, so directly the skeleton of the course. And beginning with the content generation with a video script, not yet the video, but the script of what's going to be said in the video. and lab exercises.
so challenges here functional challenges actually learning is a very deep area and well we just learned about bloom taxonomy if you don't know about it it's a type of lexicon language in learning sector to well have different level of remembering understanding applying etc but this gives a big restriction on what we can apply and how we prompt the agent. We had to respect the Chicago style grammar and a different type of question, close and open. And the Bloom taxonomy also influenced on the question.
We're gonna try the demo. And now Showtime, well, as I was saying, oh, this was only deployed today.
And well, I don't want to show you my presenter notes, but I'm doing it, not on purpose. So this is the platform that we created. And it's not going to be easy, actually, to do the demo like this, but it's fine.
You can see here that we have a library of, so we have single sign-in. So yeah, external course, this is to load the library of their courses. So they have courses on SAP, maybe Power BI, SAP for HANA.
As I was saying, the first work package we worked on was the AI generation of assessments. But since we just delivered the course outline, we're gonna go ahead and try to create a new course. So for a new course, well, I would decide what would be the title.
And so the title, I'm like super lost. Okay. JDBC for newbies.
Okay, we're gonna try to have AI regenerating our title and maybe make something that will be a bit more professional. And if it takes time, it's the connection at Critical Tech Works. It's long. So it's regenerated with another title that sounds a tiny bit more professional.
Here it's going to ask us about our audience. So here we will say for developers, again, It's gonna try to rephrase the target audience and actually tell us who's gonna be.
And then learning objectives. So here, what they will learn, actually I'm not gonna create that. I'm just gonna ask the tool to tell us what can be the learning objective.
So learn, understand the fundamentals of the Java database and learn how to establish connection. Master the creation of SQL statements. Okay, let's go for it.
So, well, could take a bit of time, but as we see here, it's showing again that it is generating a course outline for JDBC for beginners with an audience target for beginners and level developer with learning objective of understanding, etc. I will go maybe in the course here that we are creating because I can find our course that is starting. Here, this is the course outline.
So in the course outline, well, for this context, we can work on the title of the module. We can read them. We can introduce in the module.
So this was the lesson. And in the module, we can put a different topic. I can show you here that different topic we can add text, documents, lab exercise, assessment or video scripts.
Here we also wanted, our client wanted to create a landing page for every course because this they would have the opportunity to export it in Coursera for example and they would need to create a landing page for the future students to see. how great is the course. Maybe you will not be able to see it right now.
In terms of contents, so I want to go here into the video script. I'm going to try to see, well, connecting to database. So I'll try to move on and create a new video script.
And if it takes too much time, I will not. I'm speaking too much. So intro to a setting up. OK.
And now I'm saving my progress. Yes. OK. And then I can directly go on the video scripts which is being generated.
Or not. OK, back pocket. I'm going to go on something that has already been created and show you generative AI for AI educator.
So here. I can see the same again, the modules and the lesson. And here, a video script that already has been created.
Ah, it's terrible. What's happening? I'm trying to find help. Thiago, help me.
And since we have limited time, I will not lose your time in trying to make it happen if it doesn't want to happen. So it was working. Actually, Tiago told me the last response is, it works on my machine.
Sorry, yeah, you should see the, well, the titles that are created for the course and for the video script and the video script. was pretty nice to show you, actually, because it is, ah, you see? I'm a bit technical.
So here you see what is the different scenes that are created for a video script. And in the video script, well, it is creating directly a voiceover. which hopefully we can hear somehow.
But the model is generating directly what is said in the video. The images, it is proposing some images, and we can regenerate images. You see, like this.
Welcome to the module titled Identifying Use Cases, Applications in Education. In this section, we will delve into the myriad ways generative AI can transform educational practices at all levels, from K-12 classrooms to university. So you understood the idea was actually to be able to generate the video script and the audio also.
Well, the most easy thing to show you was the AI assessment, but I chose the hard way. That is the one that we actually put in production today.
The AI assessment was something which is generating pre-prompt questions of multiple answer, multiple choice question for the content creator to be able to create their own assessment without thinking about the question. As you can see here, you can give hints to the model, feedback, negative, positive, and approve some questions.
So it will be it for the presentation of the tool.
Not such a big success, but I will move over with the rest of the presentation.
OK.
So the conclusion and the key takeaway for this project was that actually AI projects is a typical dev project and AI is only a tool. Data confidentiality constraints have been very, well, big constraint actually. And it's not infallible.
AI assisted is the key. It was a lot of fine tuning and actually discovering of new models along the way. As you can see, we went from trying all the models and this is what was important in this work.
On this first assessment generation, well, the very, the best success for the client was that normally for creating 10 questions on a module, it would take them two hour and 30. And well, when the tool function, it actually takes them less than 15 minutes, just one minute for generation, and then they just have to review and interact with the tool.
So that's it. Thank you.
And here is time for questions.