All right. How many of you have ever wondered how Ferrari is able to build such cool cars and also deliver exceptional performance and unmatched um power to their demanding customers and Autodesk. How they've been able to innovate with generative design more than eight years ago. And what are they doing today with generative AI? You will learn about all these topics and much more during our breakout session.
My name is Alex Francois-Saint-Cyr. SaeR. My pronouns are she and her, I am the business development head for product engineering and development in the automotive and manufacturing business unit in AT, at AWS.
So first I will discuss the engineering and product engineering and development challenges that customers like you are sharing with us. Next, I will dive into how AT AWS. We embrace these challenges and tackle the ever increasing complexity of manufacture products. The best way to experience what we do is to hear it directly from our customers. So Mao and Giovanni from Ferrari and then Heather from Autodesk will give you insights on what they've been doing and building uh with DBS services on the cloud to better serve their customers.
Let me now dive into the product engineering challenges. So the first one is about performance and scalability now with the increased number of product features and the increase also of like focus on higher um quality engineers need to have the ability to work on more complex interconnected user cases. And in this case, they have to do really a combination of multiple disciplines such as mechanical, electrical thermal and embedded systems.
This challenge has led some of the mid-sized manufacturers to look at applying a child development, not only for software development, but also for physical product, hardware development.
Next on the list is data friction. We hear from you that you are experiencing a lack of data flow across all the business process processes regardless of the source format. Now to get the most out of product development ecosystems, manufacturers need to be able to seamlessly interconnect multiple engineering platforms and applications so that they can link key activities such as a requirement, design analysis, verification and validation.
Finally, the last major challenge is global collaboration. So manufacturers are reporting to us that while their teams might be able to um use a single source of the truth um in digital format. Well, what happens is that sometimes it's very difficult to extend this to any stake, shareholder stakeholder, regardless of their location.
So to tackle these challenges, manufacturers uh started uh look using what we call model based design and engineering approach. Whereas before they would have been using a document based approach.
Now, typically, the starting point with this approach is a model such as a three dimensional geometrical representation of the physical product. But that's typically um defined using computer edit design or CAD that's ACAD model or you can also have simulation representations commonly defined using computer eded engineering or CAE uh applications. The ultimate goal of using these is to have models that are not only full depiction of the physical product, but there are also highly interconnected so that they mimic the performance of this product. And those typically are, you may have heard of these are called as digital twins, but you also need to make sure that this data is also seamlessly interconnected and therefore you're able to create a digital thread for across the product life cycle.
Now, I have a question for you, question for you. Where are you on this journey? So do we have anyone in the audience that has a model two, which is where you have uh a level two, which is where you have models that are being shared as part of the uh design uh deliverables. Anyone? Yes, great. Ok. That's great to hear. Thank you.
Um now I'm gonna ask, do we have anyone that could build digital twins of their products yet? Much more challenging? Right? Yes. So this is where AT DBS where we can help.
So during the life cycle of a product, uh you see a manufacture, manufacture, product going through uh multiple activities uh starting with strategic planning and ending with recycling. The main activities that are related to product engineering and development are product planning, product design and validation and stimulation.
Now, of course, these activities generate a lot of data and that data needs to flow upstream and also needs to flow downstream to ensure traceability. And so that a single source of truth in order for the product is generated, the data generated from uh production or even from service also needs to be able to move back upstream to the product engineering and development teams so that they can continuously improve the product.
Now, that's why it is critical to build a close to digital thread. And as Peter Bilello uh mentioned that he's the president and CEO of SIM data, he mentioned recently in a webinar on product life cycle management that cloud is the technical foundation for the digital thread.
Now diving deeper into the product engineering and the development activities, we identified six specific areas where AWS and our partners play a key role for our customers. For virtual design. We have a service such as an IDCV, which provides a high performance graphical user uh experience for remote desktops. While um Research and Engineering Studio, a newly, a newly released product provides a way for uh to to leverage multiple services such as Amazon, EFS and Amazon um OpenSearch to help administrators in terms of like managing and deploying a fleet, a large fleet of virtual desktops instances while keeping a very close eye on the budget.
When it comes to CAD and product data and life cycle management, we work closely with partners such as Siemens PTC and Auto Auto desk to ensure that their applications leverage AD BS services to optimize their capabilities and continuously add new features that better serve their customers. Heather will have the pleasure to actually provide you with more insights on the different A BS services that Autodesk uses in order to create innovative software solutions with regards to CAE and um validation. Abras parallel cluster is an open source um source cluster uh management tool and that allows to quickly deploy and manage high performance computing clusters. You will hear from Giovanni, he will take a deeper dive into how Ferrari is leveraging this offering for faster virtual prototyping.
We also recently um released a product called Twin Flow which allows you to uh build and deploy predictive models at scale. Now, we have additional, you know, key technologies that uh are we are helping our customers with when it comes to uh product engineering and uh development, for instance, setting up a secure global collaboration so that models can be seamlessly uh shared but also in a very confidential protecting, you know, in intellectual property using Special Review in order to leverage and democratize 3D and augmented reality, fully like to get a full experience across the engineering and development processes. For example, as you want to do some collaborative, collaborative design reviews and applying AI and machine learning um to capture knowledge in order to assist users predict outcome and also generate an intellectual property.
Now, obviously, AI ML and JAI has been have been on everybody's mind. I don't know how many times we've heard AI, you know, the past few days, right? So um what i like to do now is actually take a deeper dive with you to show you what you know, we could do with those technologies when it comes to product engineering and development.
So i don't know if any of you attending on Monday, we had a session on AI powered uh manufacturing digital thread. Anyone attended this, I'm not sure. Ok. So if you haven't, um this was a session that really looked into how you could uh interrogate the product life cycle, you know, of a of a specific product. And in this case, look into data associated with, let's say a defect, you are specifically interested in, in, in, you know, looking at a defect. And then you could actually say, ok, i need to understand what supplier might be able to fix that defect, but i need to make sure that i know what kind of quality, you know, this this supplier can also provide for me. And at the same time, you could also look into any of the supply chain, you know, areas factors that you need to consider. So we do have that demo, you know, running at our booth in the WSWS industrial. So feel free like tomorrow, i think at noon, we definitely have it. So feel free to stop by if you want to see it
Cadence, a leader in electronic systems design and S partner recently mentioned that now the users of their applications can, can design printed circuit boards five times faster than before by using Gen AI and AI uh as part of their applications, you can also explore the new uh car, new car designs and optimize the aerodynamic performance of this car using JAI and computational fluid dynamics. All of this just from a single car, you know image.
So these capabilities you can see create tremendous benefits for product engineering and uh development. For example mckinsey estimated that the productivity will increase at a value of 10 to 15% of the research and development cost. You can also envision how this will impact the way the rate at which innovation happens and how it will reduce the risk of recalls while increasing product quality and finally, how it can enhance the way products are manufactured using new and sustainable materials.
So now G Giovanni and Mao will take over to share with you how BS helps drive product innovation and enhance customer experience at Ferrari.
Thank you so much.
Good afternoon. My name is Giovanni Longobardi, the Cloud Operation Manager for Ferrari Digital and Data department. Before starting with content session, we would like to introduce you how we tell our story of passion, how we capture our constant desire to push boundaries. Let's have a look.
[Ferrari video plays]
Fantastic. Our passion lives on emotions which transceive the road and track. It sets ambitious targets and expectation to match tradition with innovation. It is required to feed the ability to provide revolutionary technological solution with exceptional artisanal craftsmanship which allow us to create icons that stay timeless. Our talented people are our greatest resource, but they can pursue the extraordinary only by working together as a team. For this reason. We have founded some guiding principles which will be introduced by Mauro.
Thank you, Giovanni. Hello, everyone. Nice to be here. Uh my name is Mauro Coletto and I'm leading the data science and business analytics team uh in Ferrari. So we have seen from the video, it was clear what's Ferrari, I think it's very well known, but Ferrari is racing a sports car and also lifestyle and we are basing our strategy on technology and innovation, how we do that. Uh so we are following some very relevant principles that are displayed in the, in the slide.
So the first point is continuous learning. So we all focus a lot on learners. So learner for all Ferrari people, it's very relevant. The concept that was introduced is confident humility. So by confident humility, we mean we know what we we know, but also we are open to learn always more. So um on this point, uh we are working a lot. So there is a, this is a world where these things are changing very fast and technology is changing and moving very fast. And we have seen the cloud transition and everything that is happening with the AI and JAI. So it's very relevant to, to learn all all this stuff. And that's a very important goal for, for the company
So we have been organizing different uh stuff within the company. One event that we we we do, it's uh for all the employees and we call "Data Driven Days". During these days, uh we teach to the people what, what is data, what's the value of data? What's uh uh how can you represent that to data vis validation? And also a little bit of machine learning.
We also do some session for the business which is very relevant and we work with AWS for those sessions. And during those session with business owners, we tell them what's machine learning, how they can apply machine learning in their use cases. So uh this is very important also to bring new use cases.
Uh within the company, we also have uh some technical session, of course for data scientists, uh data scientists are spread all over the company and we do specific training on machine learning and uh and AI. So that was the first value. So the very first value we always learn so continuously learn, which is very, very important in Ferrari.
The second value is uh partnership. So we learn, we know we know a lot internally but also we rely on partners and on their knowledge. So of course, we we can mention AWS as an uh very, very relevant partner. Well, we have a really a lot of partners with whom we work every day to improve our technology and uh and innovation.
Third value is collaboration. So by collaboration, i mean, specifically internal collaboration. So what we have seen that data science and machine learning skills are spread all over the the company. And that's common in big companies, medium and big companies. And so what we organize is what we call "Data Science Hub". Data Science Hub is a meeting that we do re regularly with all the employees. And the goal is to have the occasion to train the people to explain projects and uh to build a virtual community.
So the community is not in the organizational chart but the idea that we have the data sciences spread in the racing part in the Scuderia Ferrari, we have some of them in the research environment, we have some in the digital team. So the idea is to bring all of the people together and to work uh together uh creating connections.
We have been touching different topics in the recent sessions. So this is uh an example of the last uh topics that we we touched. So what about these topics? So they have all in common data. So for those who were in the session in the keynote this morning data was the initial part and that's uh the basic uh part for, for every journey, let's say.
So, uh data together with people and innovation is real crucial, really crucial value for Ferrari and for product innovation in product innovation, we really rely on data and we have a data driven approach. And in the last months, we worked, uh we have been working really hard uh in our department to rationalize our data sources and our data lakes and we did that because we want to accelerate our multi cloud journey that we started already.
So here you see our Ferrari Federated platform. So this picture shows uh what's what's in Ferrari and all our platforms, but we will focus specifically on vehicle platforms. So, so in particular, we have develop two platforms. So one is be do platforms because they do a platform as in and it is in AWS uh uh technology. And the second one is the Connected Car platform.
So Connect Car platform is the platform where we have all the signals coming from the cars and uh it's separate for the vehicle platform for security reasons. So we want to keep uh segregating the data for security reason.
We also have the Vehicle Data platform that i'm going to to explain right now. So we'll focus on the Vehicle Data platform because it is a very, very good example, how we can provide a lot of information to the engineering process.
So as you see in the slide, the goal of the Vehicle Data platform is to collect data from different data sources. We have uh data sources that are legacy systems, legacy systems. So for example, um our diagnostic data also all the data regarding the product life cycle um and also data coming from our content management system and enterprise relations relationship um platform.
So we have all of this data that are based on vehicles, all information regarding the vehicles, but also we have unstructured data. So for example, we have all the data coming from prototyping and testing. So all this data, they go into a platform.
Um and uh the, the, the technology based the technology uh for this platform is DynamoDB. We also use uh uh AWS RDS APIs and also the S3 storage. So all the information go in the platform and how we serve this data. So we have 22 options.
So first one is through APIs. So we developed the standard APIs that uh uh to deliver the data to all our applications and the application, i mean uh engine engineering process, but also all our digital channels. So websites and apps and also social media and all processing that are related to the vehicle within the company.
We have also another stream of data which is the um reporting data and machine learning um stream. So we use this data also as an input to machine learning models and AI and i will describe in a little uh what what is going on and also we have a lot of reports and dashboarding.
Ok. So i'll, i'll pass back to Giovanni that will get a little bit into what is computer aided engineering process in Ferrari.
Thanks ma so now with this guy k tools allow engineers to vitro simulate uh the product or components response to load conditions in the operating environment, which means that teams can spend more time in the product development. However, leveraging kai was not enough due to continuous increasing computational resource demand.
Other challenges had risen some of them. How can we improve our scalability for i peak workload management? How can we reduce the computation of solving time and the us user experience? And how can we reach these achievements by keeping our t ceo low as well?
Here, you can find which type of simulation are possible. Thanks to k on the left side, you can find a typical sloshing simulation which predicts the fluid motion inside the typical oil tank. On the right hand side, you can find a typical gas engine exhaust simulation.
So even if chi gives us the opportunity to solve typical engineering problems like structural behavior prediction and fluid mechanic simulation, it was required to to rethink another approach regarding how to deliver the computational resources to our engineering teams.
He here the cloud comes into play. Let's see a typical application field for chi which is the i a performance computing or hpc. Our hpc hybrid architecture in the AWS cloud leverages some cloud native services. Let's see some of them.
First of all direct connect between AWS and Maranello for high speed performance uh network.
Second AWS Parallel Cluster, the core component of our arrival architecture with its compute nodes for calculations and the node for orchestrating them.
Third one, the RDBMS service in Amazon Aurora for keeping track of our job scheduling history.
And finally, Amazon FS six for last for an e performance NFS sharing plus Amazon S3 for data sharing and keeping our storage costs low.
But before proceeding with a productive scenario, roll out, we have exacted a huge benchmark test campaigns by comparing the computation of solver time obtained uh by leveraging the hybrid architecture, we once obtained it on non premises only cluster.
As you can see each type of simulation has shown better results in terms of computational cycle time saving. Moreover, during last month, we have also introduced some optimization in our overall architecture. Let's see some of them, these benchmarks have been obtained uh basically by working uh and optimizing three principal components.
First one network. By enhancing our data transfer performance, we have improved the 75% of our use cases and user experience by introducing new u weeks technologies for simplifying pre and post processing models. We have improved the ni 91% of our use cases.
And finally, scalability thanks to virtual unlimited resources provided by the cloud, we can avoid the on premises cluster resources saturation, which means that we have improved the 100% of our use cases.
These benchmarks encourage us to considering introducing a new type of experimentation and we will investigate on this. In the meantime, let's see some machine learning use cases and other potential application field back to tomorrow.
All right. So we have seen the the data layer. So what is on the basis we have seen our hpc technology. Let's move a little bit on the application uh layer. So we'll focus specifically on machine learning case use cases which are more interesting.
So uh before going into the use cases, i will describe, describe a little bit the architecture. So we'll focus on the vehicle of the platform that was the platform that i described, which is the input for all the use cases. All the machine use cases.
We do use the SageMaker environment. And we also use all the of the shelf algorithm together with custom algorithms that we develop uh uh internally. So most of the use cases that we work on are about the recommender systems are about anomaly detection are about computer vision and forecasting regression.
So we do have a lot of use cases, but i'll focus on a couple of them just to give you an idea what we are doing with this uh infrastructure.
So the first uh use case uh is about uh virtual sensing. So we use a lot of these. So that's an example for sports car, the department uh business area. So we try to predict the value of some relevant uh uh metrics within the the the car within the vehicle uh starting from other additional uh metrics that we have.
So for, for example, in this case, we want to predict uh the friction coefficient of the car starting from value that we have like temperature in a part temperature, temperature, extra, external part temperature, uh speed, uh uh torque and so on.
So we developed a model, a custom model in this case that was able to predict uh the value of the friction coefficient for the car starting from all these values. And the prediction was very, very accurate.
So by adding this information, so by adding the friction coefficient, we can improve um um the the braking system based on the the style of the driving style of the of the customer.
Um let's move on. So this is uh one of the general use case. So we do a lot of this use case for diff different metrics.
So we do a lot of the CS. Uh, but also like uh I want to, to, to show this other use case, sorry. Ok. So that's to complete uh the the loop.
So uh to show you all the information that are come that are very relevant for the engineering part that can come also after the delivery delivery of the car to the customers.
So in this case, uh we see a computer vision uh use case, machine learning computer vis use case where we try to auto label uh pictures and all these pictures can be picture that we find uh online. So web content and we developed uh uh a custom model in this case, uh inst maker using recognition and also custom uh label.
And so starting with the identification of the object. Of course, we can say if in a picture there is a car and if there is a ferrari and also we can say, which is the specific model of ferrari that is in the, in the picture. So we can tell if it's a ferrari portofino, if it's a ferrari roma and the accuracy is more than 95% we can also say who are, who are the people in the picture.
So uh if the beach, the people are celebrities, we can of course recognize through the of the shell algorithm. But also we custom uh train the, the, the model with of relevant ferrari people that are uh for example, our chief officers.
So in this way, we have a lot of information about the content that we can find also externally. And of course, this is important for archive reason reasons. So in order to, to pick up content to see what's the content that's uh outside. But this is our information that can go back also to the engineering part.
Uh of course, uh uh we have some limitation on privacy and we follow the gdpr constraints. But uh if uh we can, we can provide this information to the, the engineering part that can have the feedback from users from customers on the real product.
Let's say last point, I want to touch, of course, is uh generative a i. So all the conference uh is about generative a i. So we know that it is a big potential. And of course, also in ferrari, we are testing and experimenting a lot with generative a i.
So what's uh what's generative a i and hawaii is so important, it changes a lot uh the speed uh the speed that uh needs uh for, it's important for, for, for engineering, the for process engineering.
So usually when we have to design apart, uh we go through a lot of uh runs a lot of cycles that can be weeks to design a component, a specific component with generative a i, we have the possibility to have a lot of uh different variations of the same component can be after judged evaluated by the, the engineers by the designer and can be selected for prototyping prototypes.
So that's a really very important uh game changer and we are working a lot also on that. So I will now leave the floor to ether. She will go even deeper on the general tv r part uh from outer desk. You, you thanks now.
So hello, I'm heather carrick and I'm gonna talk to you about how auto desk and aws are working together to help wrangle some of the complexity in the manufacturing industry that alex talked about. And yes, I will be talking about generative a i.
Uh so by show of hands how many of you have heard of auto desk? Oh, ok, good, great. We got a lot of, a lot of uh fans in the room. So for those of you who didn't raise your hand. Autodesk makes software that helps people design and build really everything around you. We span multiple industries, architecture, engineering and construction, product design and manufacturing, which is my department and media and entertainment.
So if you think about the plane, you flew to get here, the movie you watched on that plane, the car you rode in to get here, the building we're in and the chair you're sitting on. All of those were likely designed with auto desk somewhere in the in their workflows and just for some stats, you know, autodesk is about 40 years old. We have 13,000 employees globally and over 6 million subscribers to our tools and those tools create chairs and planes and and buildings that impact billions of people around the world and underpinning all of that is aws we need a connected technology stack to connect things like design and compute to create workflows that flow within and across all of these industries.
A prime example of this is with autodesk fusion fusion connects all of your tools in the manufacturing space. So design simulation, documentation, manufacturing, electronics rendering, you name it all in one platform, we started fusion because we saw that you needed a single environment for design and make workflows a place where data can move easily for different kinds of teams and people to work together um across different systems and locations and partnering with aws.
We've been able to optimize how we leverage the cloud so we can offer this tool with really sophisticated workflows at an accessible price. So throughout the session, we've talked about some of the complexity and the uh the challenges that the product life cycle is facing. And this diagram is really the one that autodesk is is working on.
So uh you see there's all sorts of people and all sorts of softwares and vendors and systems and file types. Anyone see themselves on this diagram by any chance person up at the top person down the corner. Uh this is how manufacturing works today and it's kind of bananas frankly. And we think at on us, we think that your software tools, your design tools should really help smooth all of this out, make this this easier to understand, especially because there are a ton of external challenges facing manufacturing.
There's a labor shortage, there are supply chain challenges, there's an increasing need for sustainability. And what often happens is the person way off in the top left corner makes a change, not realizing that it totally breaks what the person down in the bottom right corner is doing. There's all sorts of multiple variables in, in different ways they interplay. And it's really hard to understand the impact of all of those things.
And so we are using tools like autodesk a i tools to really help wrangle that complexity. So we have auto desk a i tools in fusion and in our other software across the autodesk design make platform.
So we're a software company in order to create these a i powered tools, we require a lot from our technology stack and it's the kind of things that we get from aws. We need high performance, we need high performance compute in order to do a lot of this computation, you our customers trust us with your business. 6 million people, trust us with their business. We have to keep your data secure and we have to offer our services to you reliably.
And as we're seeing very acutely this year, technology is changing all the time. And so we need to be flexible and scalable to absorb new innovations and scale to, to greater and greater sizes and and further scope.
So uh one example of how autodesk and aws have partnered in this way is with generative design. So we started generative design about eight years ago in 2015 when we partnered with airbus on this bionic partition.
So i'm gonna clarify my terms a little bit. I work with scientists and researchers every day and it's really easy to get hung up on semantics. But I just want to clarify this generative design is different than the generative a i that got super popular this year. Both are generative as the name implies. Uh so they create new content. But generative a i is based on machine learning. If you train machine learning models, you show bazillions of examples, train models that learn patterns, those models based on what they've learned will create new content.
One thing about generative a i especially right now is it's kind of a black box. It's hard to always know how you got a particular outcome and people have probably heard of hallucinations. It's possible that the suggestion offered by your generative a i model might look kind of cool but may actually be false in some way. And in the manufacturing industry, that's, that's kind of a a challenge.
Uh so generative design by contrast uses equations. So it's all deterministic. There are simulation equations and physics equations. And so as a user, you describe what you need your part to do. What material is it made out of? What weight does it need to bear? And then under the hood, a bunch of different equations get crunched and multiple totally valid mathematically accurate solutions are computed and offered to you.
And because we know all the equations, we can see the impact of different variables. And so what it offers to you is the ability to understand those tradeoffs to see what happens if you, you know, try to do this faster. What does that do to cost if you try to make it lighter? What does it do? To strength and so you can actually see the interplay of those variables.
So one example of this is general motors use generative design and additive manufacturing to design and build a functional and lightweight seat bracket. Their goal was to optimize for weight or mass technically. Uh so in the past, human designers would have made two maybe three designs for this challenge. But those human designers partnered with our software made 150 viable options.
We actually had to train a machine learning model to group these options to make it possible for users to navigate these choices. Ultimately, they designed a seat bracket that was 40% lighter, 20% stronger and consolidate, consolidated eight different parts down to one part. Consolidation really is helpful for the assembly process and making a part lighter and stronger is obviously better for the life cycle of that vehicle.
So aws was essential to this tool in order to compute and generate hundreds of designs and display them in a way that someone could explore that design space required a really sophisticated backend. So we worked with aws and we created a back end with parallel compute that uses both gp u and cpu. It would not have been possible to make generative design without aws s tools.
So that was back in 2015. So over the years, we've kept, kept expanding generative design. So with the fact that we have all these a s components, we can kind of mix and match and recompose new workflows using some of our old tech and adding in new technology. And that's how we expanded generative design to include things like generative fluids and automated modeling.
So, generative design is great for structural components, but i mentioned it's simulation based and if anyone has anyone used a simulation tool anyway. Oh ok. I see some hands you, you can tell your it's a, it can be a bit of a challenging environment for people. It's a really different kind of user interface than, than a design space.
And what we found was that uh not everyone could found generative design for structural components as accessible. And so we created the automated modeling tool to put generative design in in a design space which is more comfortable for some users. Also...
Um over the years, we've expanded generative design to cover multiple manufacturing methods. So the GM example used additive manufacturing but additive manufacturing is not really the best manufacturing method or appropriate for all use cases. So we needed to, to add additional equations and additional capabilities to generative design to, to be able to offer manufacturing methods that were relevant to all of our different use cases.
So now generative design, shove those more equations in and it can support uh additive die casting, three and five axis milling, 2.5 axis milling and two axis cutting. And because we have all those equations, we can really help um users understand how their choice of manufacturing method impacts their decisions.
So uh MJK Performance creates parts for performance motorcycles. They use Janita Design to design a triple clamp. And so there are three designs that they compared to the human design in the center. So you might be able to see depending on the text. The additive design was by far the lightest, but based on their production volume way too expensive. So the 2.5 access design gave them was 23% lighter and worked with their production volume. It was a, it was a more manageable cost. And so using generative design, they were able to go from an idea to a computer model to a physical prototype in their hands in a matter of hours.
As I said before, the underlying building blocks of generative design use AWS and so we can componentize them and mix and match. We do a b testing a new stuff and it lets us continuously improve this tool without disrupting any existing user.
So in our most recent update, we improved our structural design solver. Uh Mao and Giovanni talked about solve times as well like it's, it, it's a, it's really important to be able to, to be as fast and efficient as possible. With this. We've added, we now support those six manufacturing methods and we've done a lot of work to increase the robustness and resiliency of our solver.
Um and so we use in u in order to use, in order to offer flexible business models, we have to use our computing resources really effectively. So we use AWS for a mix of GP us and CP us, we use batch and step functions and we use spot instances to use exactly the right amount of compute at the right time so that no resources are wasted. So that's how we use um AWS to to really handle the multi variable aspect of this problem.
But another part of this problem is all of those red arrows. People spend engineers spend 30% of their time in these red arrows. They're doing non value-added tasks, tedious work, transcribing really data from one format to another. And they tell us they, nobody wants to spend their time that way people would much rather spend their time on the really interesting creative challenging problems, not reformatting something, telling one computer how to talk to another computer.
So uh drawings is a great example of this. Manufacturing drawings are super important. Basically, every manufactured part needs one. It's critical to ensure that that part is created as, as it needs to be, that it has all the right dimensions in the critical places. But these are really tedious to make they can take out and all they are really doing is illustrating information that's already in the 3D model. And so it's it requires a bunch of time and a particular domain expertise, but it's not particularly creative and, and folks that do it would love to do it faster.
And so now users can request automated drawings. So this is coming out in the next build a fusion. You take your assembly, click a button that assembly gets sent to the cloud with AWS technology, it gets churned and then you get notified when your drawing is ready. So in this case, this person now has 24 sheets automatically made.
Part of what makes drawings really hard to automate is that there's no one right way to do it. It's sort of half art, half science, depending on your manufacturing method, your company's standards, you may need the drawing to be set up in a different way. So we offer people a menu sort of a number of recipes. So you get to control how your drawing ultimately looks. The dimensions get placed on the sheet. Everything is fully parametric and editable. So you can still make your fi finishing touches, move a dimension here, add one there, bada bing, bada boom. But now we've saved hundreds of hours of time. So this drawing is is fully editable and we're we're really excited about what this means to free up, free up everybody's time to do more interesting problems.
So, generative design and automated drawings help superpower our customers to achieve their goals faster and both of them, as I described, use sort of rules and equations and heuristics under the hood, but there are some problems that we really can't solve that way. And so that's where machine learning comes into play. We are leveraging machine learning to develop more AI powered tools for those challenges. And we need dedicated tools as a software company in order to do that.
So over the last few years, we created the Autodesk Machine Learning Platform for all of our internal machine learning teams to access the tools that they need. The platform is built on top of AWS and it uses Amazon SageMaker, Batch, Step Functions and other AWS components to optimize how we use our resources. As we scale our machine learning footprint, it becomes essential to manage things like compute and data storage effectively.
The platform has different tools to support different part of the machine learning life cycle. But it kind of chunks out to things around obtaining and managing your data, training models, deploying models and then monitoring how they do in a production environment. With this platform, it's easier for all software developers to work with machine learning models effectively.
So for non experts in machine learning or standard software developers, it makes the off the shelf models easier to access. So folks can grab these models and plug them into their workflows really effectively and quickly. And it enables the machine learning researchers to access increasingly powerful compute, to develop and train new models, create data sets and do things like training on data without copying or moving it. When you have really large data sets, that kind of thing is really important. Um and also with data governance ensuring that that data stays in a safe place. And this platform lets us apply these governance rules to ensure that we can control who has access to what data and where that data lives.
For the past five years, the Autodesk AI Lab has been advancing the state of the art for AI and machine learning in two D and 3D geometry. The lab has published over 60 papers and several data sets for the larger academic community. Three data is a lot harder to obtain than things like text and images. You can't really scrape the internet and get a, a useful meaningful data set of this kind of data. And that's been a real challenge for the the research community in academia to to really get off the ground with in the same way that has been possible for, for data like images and text.
But luckily Fusion has an amazing and passionate user community. Over the years, many designers and engineers have published their designs to a tool called the Fusion Gallery. And so a few years ago, our AI Lab was able to pull designs off of that gallery, curate it a bit and create the first data set of its kind that had really, really rich data about how CAD models are made. And what was going on in the, in that data and we published the Fusion Gallery data set and we've added a few more editions to that data set and shared it with the academic community. And it's really been great for, for helping expedite innovation in this domain.
So building on the work of the AI Lab, we are exploring applications of generative AI and how it can further help us tackle the challenges in the manufacturing industry. One of the areas that we're really looking at right now is how to get a design out of a designer's head and into a computer. For example, on research partnered with Kia to build a generative AI tool for concept generation for wheel designs. They used AWS to generate the data that they needed and created a model that understood these intangible concepts like dynamic and photos and could also understand design principles like symmetry.
The tool enables Kia designers to quickly generate and modify their wheel designs and then ultimately create a 3D model going further in the automotive space kind of the theme of the day. Um Autodesk recently acquired Blank AI. This tool allows any automotive designer to encode their brands specific design DNA and then generate and tweak 3D designs that still kind of have the essence of that particular automotive uh brand. The designer can make adjustments using semantic controls again, you know, dynamic, unique, bold uh and immediately visualize the outcome in a rendering environment.
So we're hard at work on boarding this, this new capability onto the AWS platform and uh using AWS to help scale it to more to more users. So we'll keep working with AWS uh to continuously improve and expand these workflows. Uh so I'm really excited to see how folks like you can use ASK I today and, and the new tools that we, we will continue to share with you. Uh I'm gonna pass it back to Alex to, to bring us home. Ok.
Thank you, Heather. So, thank you very much. Um I hope that um uh what Mao uh Giovanni and Heather shared with you today made you better understand the value that AWS cloud, you know can bring to product engineering and development. Marrow and Giovanni described um how Ferrari is able to increase the innovation and collaboration with the customers and suppliers via their digital platform and also by leveraging high performance computing. Heather showed us how Autodesk uh helps their customers uh not only reduce time to market but also how they can shrink the number of designations they do through that cross functional collaboration. And by leveraging AI now just visualize how you could design an aerodynamically um improved car from just an image by using JDD I and also uh by using machine learning and how your team could potentially look at that design and experience the styling of this design of the new car via a collaborative augmented reality visualization.
So it will be more to come on that topic um early next year. So stay tuned. So thank you very much, you know, for attending our session today. Um Mao Giovanni and Heather. And I really hope that you enjoyed the session and that you were able to learn something that, you know, meet and that, that you might be able to use, you know, in your, in your day to day life, you know, as a, as a designer, an IT person or you know, on, on that.
So as you know, we are data driven at AWS. So as such, your opinion, great matters to us in order to make sure that we always improve, you know, our efforts in providing you the uh the best content. So please take a moment to fill out the short survey that we have on the mobile app and uh so that we can really, really hear from you and make sure that, you know, we always, you know, satisfy what you're looking for in terms of our delivery content.
Thank you so much. Thank you.