From the event: Mindstone Paris June AI MeetupFrom any document to a complete gamified learning path: how AWorld's AI builds education that actually sticks
View event

From any document to a complete gamified learning path: how AWorld's AI builds education that actually sticks

Introduction: A Gamified Training Experience

You have the phone in your hand I see so now you have to scan the QR code is mandatory unfortunately so please scan the QR code choose a username and put your email because we're going to play together so you will see question comes out from your phone the moment we I change slides so please keep your the

the phone in your hand and you you will see the question in real time so first

Why Mandatory Training Doesn’t Stick

question you don't have to answer to this question but it's just to ask how much do you still remember about your last work training mandatory work training do you really remember what your last work training was about so how

How many people really skim or ignore the training they are forced to do? Just guess. So apparently the majority say 49 % of people skim or ignore the training. And you're right. so 49 % of people skim the mandatory training that's not very good

half of the company is skim the mandatory training there is another problem the other problem is that we tend to forget very fast so half of what we learn is forget in half an hour most of what we learn is forget in one one day.

And this was in 1885, and this is the same in 2015. So maybe more scholars, the people are more educated, but we still forget half of what we learn in a day.

Attention, Engagement, and the Leaderboard Effect

Oh, there is a leaderboard. Who is Emanuele and Filippo? So, you have a leaderboard now.

so you see that now you are following the panel you are following the panel and desire a kick in your mind that that say to you that you have to follow because now you are in a leaderboard and so if you miss the next question you will be you will go down into this leaderboard let's move on so we started

From Sustainability Learning to Corporate Training

We were chosen as a UN official app, so we are on the home page of the UN with our application that is AWORD in support of ActNow.

UN Recognition and Early Product Validation

And we are still there, but we started working with corporates and corporates start saying oh my people is like it's playing with your content is learning about sustainability they are very engaged with your content and with your gamification you're like a duolingo of sustainability where people really want to engage with content why you don't

Turning “Boring” Manuals into Courses

allow us to put our very very boring cyber security manual into your platform so maybe it's the time that they really read the content and so we say maybe here there is the opportunity to build another product very similar to the sustainability product where you can upload any content you want into the platform and you can build courses for the company like a

Why Gamification Works: Loss Aversion and Streaks

internal Duolingo when you lose something how much worse does it feel than you when you win something so how much worse is to lose than to win you say about five times as bad okay it's two times as bad so you really don't want to lose and this is something that as I said you're following the panel maybe it's not your topic but you're following the panel because now you don't want

to lose and if i say to you that tomorrow you will have additional point and maybe you can win this this leaderboard tomorrow by answering that additional question you will come back tomorrow to answer that question this is why the streak of duolingo has people that is in street since 10 years every day for 10 years to not lose the streak so these are some numbers about a world

Traction and Impact Metrics

The sustainability app has been downloaded more than 1 million times in 193 countries, more than 28 million actions, 80 % higher completing rate than comparing to a classic LMS.

But the number that really comes to you is that you are in that leaderboard and now you are thinking, I need to stay in that leaderboard.

How the System Builds a Course from a Document

Another question, and this comes a little bit more difficult and a little bit more technical. Our engine can turn any document, also the boring cybersecurity document, and runs in 10 steps.

Who is the boss of this AI that transforms your content into a training path? Is an AI boss or is a plain code AI?

The Orchestrator: Plain Code Managing AI Agents

It's a plain code, so a classical software without AI that manages all the other AI. And you're right, it's a plain code. So our software, so the manager, the boss, is plain code.

It's a very classical plain code. Then this plain code can manage multiple agents, AI agents. when you heard about AI agent, AI agent are single AI that are working on a single topic and now we

go through all the steps okay so the orchestrator let's say is a plain code very classical plain code like the code we were we used to to write before 2023 so in a fixed order it can retry if if it fails during the process of transforming document

into learning path, it can be paused and resume, and can build several courses in the same time.

Then the boring boss manage all these different agents.

The Agent Pipeline: Parsing, Structuring, and Building Content

So there is one agent that is the parser. The parser read the entire documents. There is another agent that is the online builder. It's the one that, starting from the output of the parser, designs the course.

There is the knowledge builder that puts out the key idea, so it extracts the key idea from the documents. And imagine that this can work with documents of 200 pages, so it extracts the idea from the documents. Then there is the course builder that write slides and quizzes.

There is the content formatter to give slides and quizzes layout and color. There is the image prompt agent to create images. The attribute builders that write the title, metadata, and tag. And there is the entity creator that checks and publish everything together.

Media and Localization Agents

Then we have other three additional agents. The one that really create the images. The agent before was only create the prompt to give to the agent that really create the images.

There is the agent that created the audio because you can also listen the content. There is the translation agent that can translate in 40 languages and then there is an additional agent that is coming the next day that

Why Agents Work on Small Chunks (Not the Whole Document)

can create video avatar of the content okay so next question is when it writes a section does each agent see the whole documents or just its own small piece okay five went through and yes

the agency only is small pieces why because otherwise he is lost if you gave every agent a 200 page document the agent will be lost in the document and you're not sure that the key concept will come out

in the course so the boring boss plan the code, set the order, not an AR, one piece at a time, each writer see one idea, not the whole file, it renders itself

if a step fails and the person checks it at the end and i can check at the end as a human operator if the course is okay if the learning part is correct

Implementation Layer: Google ADK and Agent Coordination

we use google adk we have been awarded as up of the year by google in 2023 together with chat gpt whatsapp and many other best up for good in 2023 and we have been we have been part of the google accelerator in Europe and so we start working on the Google ADK Google agent

development kit and the Google agent development kit allows us to work in real time with all these agents together so it's the layer that allows the agent to talk each other and to rent three when they fail it's very easy I

Retries, Error Handling, and Reliability

basically already gave you the answer to this question if one of the step it's a a temporary error while building a course what does the system do and the answer the correct answer is that rent is that step on its own okay so the agent keep rendering the steps so all of

Where It Runs: Cloud Infrastructure

this runs in the cloud on google part partially on google partially on aws but the ai part is on Google.

If a step fails, it tries again. If it really crashes and it happens, it can happen, it starts over from the beginning.

Very good.

The Reveal: You’ve Been Using the Product

So, where is the product? Where you can find the product? Where is... anyone has any idea of where is the product?

So, the real secret of this presentation is that you have been inside the product the whole time because this presentation is made by our everything built starting from my idea so you have been inside the product the

technology so I put my idea together I upload my idea and this is a learning path is that an a word learning path with question with gamification and with

the old time and this is the final leaderboard

filippo where is filippo you won okay so filippo 400 points you won

Wrap-Up: How to Try It

and so these are my contacts if you want to send me an email or a document and we want to try to create some gamification on the documents you can go to our website

uh this product this product is called evolve so if you If you go to the Evolve section of the website, you can activate a free trial of our product and start trying how it works to keep people engaged.

Finished reading?