- il y a 1 semaine
The smart airport agentic assistant shaping the future of travel
Catégorie
🤖
TechnologieTranscription
00:00Hello everyone, it is a pleasure to be back on Discovery Stage for the third year in a row.
00:07Three years ago I came here to present Van Gogh.
00:11It is our first project.
00:14Van Gogh is an avatar that was put in Musée d'Orsay
00:19and the purpose is to create a new way of culture mediation.
00:23Last year I went here to present Louis,
00:25which is an avatar also to assist people here in Vivatech.
00:30This year I present Lisa with César and Christophe
00:35and we will talk about Lisa.
00:39That's why I want to let Christophe and César introduce themselves.
00:44So Christophe?
00:45Yes, so hello everyone, my name is Christophe.
00:48I'm the founder of Jumbo Mana.
00:49Just quickly, I've worked on AI for 15 years ago
00:56and I have started on General TBI in 2020.
01:02And so I created Jumbo Mana.
01:05It's a startup specialized on digital characters, interactive characters for a public space.
01:12and we focus really to create the best solution for large public in public space.
01:19And all problematics concern about how to put authentic AI fully autonomous
01:27to respond to international travelers in a nosy environment like Vivatech or an airport, etc.
01:39César?
01:40Yeah, hi everyone.
01:41I'm really happy to be here to testify about what we've done with Jumbo Mana.
01:47I work for Vinci Airports.
01:49So Vinci Airports is the first private airport operator.
01:53We've got more than 70 airports across the globe.
01:57And to foster innovation, we rely on what we call center of excellence.
02:02We've got three.
02:03One in Lisbon.
02:04One in Gatwick.
02:05And the third one in Lyon that I'm leading.
02:08So I'm leading digital and innovation, customer innovation in Lyon.
02:11So focus on customer engagement.
02:14And before I used to work in e-commerce on e-channel for retail and department stores.
02:22What we want to do in Lyon as a center of excellence in customer innovation is to experiment, explore
02:29new services, new solutions to improve the whole customer journey by speeding the flow
02:40at the airport, easing the flow at the airport, improve interaction and engagement.
02:45And we'll talk about that a lot.
02:48And then enrich your experience as a passenger at the airport with new services.
02:54Thank you, Christophe and César.
02:57Now, before diving in, I want to show you a video, a demonstration of LISA and how it works
03:03in the Lane Airport.
03:05And let's see it.
03:12Bonjour et bienvenue à l'aéroport de Lyon.
03:15Je suis Lisa, votre assistante virtuelle.
03:19Si vous avez des questions sur nos boutiques et restaurants ou même sur votre vol, je suis
03:24là pour vous aider.
03:37La porte d'embarquement pour Barcelone avec EasyJet Europe est 10, 10.
03:44Au revoir, passez une bonne journée.
03:51So Christophe, after your project of the digital twin of Van Gogh in the Museum of Orsay, why
03:59did you choose to develop this technology specifically for airports?
04:03Yes.
04:05So we started with the digital twin of Van Gogh for the Orsay Museum.
04:11So we presented it at VivaTech 2023.
04:15But when we bring this project to the large public and professional, it was really a demonstration
04:25of our AI, our capabilities to humanize AI.
04:32And our purpose is to create a solution, like I said, for public space generally, not only
04:42for museums, but for airports, malls, etc.
04:47And especially for mobility.
04:50And so, we are looking for a partner to create a unique reference, a world reference of
05:00Atlantic AI for public space.
05:02And airports are a fascinating environment for that, because it's a noisy environment.
05:10they are international travelers with multiple accents, multiple language, and people are stressed.
05:21She needs information quickly and to be reassured during their stay at the airport.
05:29So, for us, the airport is an ideal environment to test Atlantic AI for autonomous, for large public.
05:39So, it started with Van Gogh.
05:42And after that, you needed to demonstrate how you can use this technology, which is called
05:46Atlantic and Generative AI in place, crowded with a lot of noise, to manage this flow.
05:52So, and you see that the airport is the best place to do that.
05:56So, my next question is, how did you meet VanCy airport?
06:01So, we met VanCy, especially Cesar and his colleague, Anatoly, during VivaTech 2023.
06:09So, they saw our Van Gogh demonstration.
06:15I suppose they like it.
06:17And when I came back to them in the last year, to tell them we are looking for a partner
06:26to
06:26create a unique reference for the mobility sector.
06:29They immediately share the same vision that us to create this new technology for the airport area.
06:41Okay, perfect.
06:42So, Cesar, from the airport operator's perspective, how do you view this challenge on the ground of the airport?
06:51Well, the objective was not to do AI to do AI.
06:59The first point was that of healing or fixing a real pain point on the field for the teams or
07:08for the passengers.
07:11And as you said, Christophe, airports are big spaces where we have to handle with a great amount of inquiries,
07:20which are very diversified.
07:22And we've got staff on the ground.
07:25The staff can be everywhere, anytime, for everybody.
07:30So, for us, as we remain an environment that can be a little bit stressful,
07:36we wanted to use this solution and AI to improve the service quality, which is one of our core concerns
07:48in Vinci airports.
07:49And it was a way to do it in a very engaging, interactive way to automate basic, not only basic,
07:58but question, a great amount of question.
07:59Because our staff always answers what are more similar questions.
08:08And it's a way for us to save time and to dedicate our teams to other more complex cases or
08:17questions.
08:17The question is not that of saving staff, but to complement all the information services that we have at the
08:27airports and to reallocate our staff on higher value tasks.
08:34Okay, so what I can understand that we are not here to replace the staff, but we are here to
08:41complete its role in the airport so that we can manage the staff to do a big role in the
08:47airport more than replacing it.
08:49That's the friction that AI nowadays creates, right?
08:53So, my next question is, can you give me an example of difficulty passengers that they face and that Lisa
09:02could help solving it?
09:06Well, just to give you figures, we've got 40% of the inquiries of the passenger question are about boarding
09:14gates,
09:15what time the boarding gates will be displayed, how to go, and they ask always the same question.
09:23And with Liza, you can ask it with voice command.
09:27You'll get the direct answer that has UQT and the videos, and you will be displayed the indoor wayfinding capacity
09:34that you can get on your smartphone to go to your boarding gates.
09:38So, if you've got big issues, staff are here to accompany you until the solving, but if you've got just
09:47one, not simple, but yeah, let's say all the other type of question,
09:54Liza is able to answer as if she were an agent.
10:00Okay, perfect. Now, my next question is to Christophe. Let's tell the audience how this collaboration, the story of this
10:09collaboration, how it works, how we did this partnership, the development of this project.
10:16Can you tell us more about this project?
10:18Yes, of course. So, the collaboration started last year. We started in summer of 2024.
10:25We have worked during six months on the project from the creation of the avatar to the deployment of the
10:33Lyon airport.
10:35And I think the real key success of the product and this partnership is that Cesar and his team, with
10:46our anatomy and other colleagues,
10:49was very involved during the process of the creation of the avatar and the AI.
10:55And we worked with a lot of big companies. And I think with Vinci, it's a base use case of
11:04real partnership between a startup and a company for two things.
11:09The first thing Cesar said, we are not here to make AI for AI. We are not here to make
11:15a fashion or a gadget like that.
11:18We all need to respond to needs. And we co-construct the solution with the employees on the field with
11:25Vinci on the airport.
11:27And the second thing is, there are a lot of big companies that did, the director said, we need to
11:34go to General TBI.
11:36We need to go to Agent TBI. I put the check and just do it. But behind the employees are
11:46not enough involved.
11:48They are not respond from the company, etc. And so we are at Jombomana, our startup expert on AI. We
11:57are expert on energetic AI.
11:59But I don't know the skills and the work of the airport. And I need Cesar to tell me what
12:08is the needs, what is the concerns,
12:11and how we can improve our technology to respond exactly to the needs. So, this is the big thing.
12:23I know, for example, for the creation of the avatar, for the visual, we make at Jombomana market studies to
12:33understand
12:34what is the correct appearance of avatar to make sense and engagement for a large public.
12:42Right now at VivaTech, we are all a fan of AI or we are educated about AI and technology.
12:50But if we go outside and you walk 300 meters, people fear about AI, have concerns about AI.
12:59So, we need to make a product not for us, but for a large public, for all travelers.
13:07So, we co-create Lisa and the avatar directly with the employees on the field of Vinci.
13:15And the staff, they don't show the avatar like an enemy, but like a virtual assistant.
13:23They will accompany them on the field.
13:28And this is the second point, that this product is not a final product that we deliver to Vinci.
13:36It's a true construction to make the right product in the right place that responds to the right needs.
13:44So, the key success of this collaboration is that the team is involved of Vinci, that they show us the
13:52real need of this solution.
13:54The second thing, it is not trendy, because when we hear about Generator BI or Agenda BI, people and big
14:00companies, they want to do that.
14:02But the key of this success is that there is a need of using Generator BI and Lisa.
14:07Now, I'm going to ask another question for you, Christophe.
14:11It is a little bit technical, but can you explain the technology behind Lisa and how do you make sure
14:19that it responds quickly, with no hallucination, with the veracity of the information and especially in different languages?
14:26Because you said that Lisa talks more than 17, it is capable of talking more than 17 languages, but in
14:32the airport now they choose to talk two languages, which is French and English, right?
14:36Yes.
14:37Yes.
14:38So, like I said in the introduction, we developed a solution to make the Atlantic AI in the front of
14:47a large public.
14:48And we need for that to create a real confidence, a real trust between the companies, Vinci and the travelers.
14:57And the AI needs to be trusted.
15:00So, the first technology we developed at Jumbo Mana is to create our own AI reasoning model to make sure
15:11there are no hallucinations and all the subject is constricted to the airport topics.
15:16We have won several prizes for that, like the iLab prize from the Ministry of Research.
15:24And we have labelled Deep Tech from BPI France for unique technology to make sure the AI is under control.
15:36And along the side, we have several AI.
15:40We have six AI models.
15:42Each model has billions of parameters from voice recognition to AI reasoning to voice synthesis.
15:53And to make sure these six AI models can respond in a few seconds, less than two seconds.
16:03We have a partnership with NVIDIA.
16:07We are a member of NVIDIA Inception.
16:10And we have access of NVIDIA engineer to make sure we can deploy our AI model directly on their GPU
16:21and to optimize the inference on the hardware to make sure we can optimize.
16:29And I have a fast and quickly answer beyond all the AI to control and to analyze the question and
16:39the response.
16:40Okay, perfect.
16:41Maybe we can show Lisa over here, the face of Lisa.
16:46And now we're going to talk with César about the appearance of Lisa.
16:50So why did you choose this DA, let's say, artistic direction?
16:55And let's talk about the appearance.
16:57She is not a hyper-realistic.
16:59She is not a human-like avatar.
17:02But it is not also a cartoon or Pixar.
17:06It is something, let's say, new and dedicated only to L'Avancy Airport and Lyon Airport.
17:14Can you talk more about this DA?
17:17Well, we talked at the beginning about trust.
17:20And Christophe, you mentioned the way that people have to be reassured at the airports.
17:28It was a real stake.
17:30I mean, we wanted the avatar to be reliable.
17:35So we choose not to go on a Pixar or cartoon-like.
17:40But on the other side, we wanted to avoid what we call the uncanny valley.
17:46I mean, if it's too metahuman or too human-like,
17:50if you've got any imperfection or any small bugs,
17:54it can create and provoke unease or discomfort.
17:58So we decided to make the choice to be in the middle
18:01and to find this DA which was for us the good solution to be human-like,
18:10but not human and not cartoon-like.
18:14One interesting point is that everything has been done involving the staff, as you were saying.
18:21We've got a core team which is really committed and staff on the ground.
18:25So the look and feel, the outfit, the content, the UX design of the interface was decided by my team
18:36but with the collaboration of all the staff which is daily on the ground.
18:42And it was really important for us to do so.
18:45One last thing, we made the choice with Anatoly, the project manager on this specific project,
18:54to start with a human-sized avatar to be as engaging and interactive as possible.
19:01So today, as you could see, it's in a big, as you've seen, it's in a big kiosk,
19:05so that the interaction is really human-like.
19:08It's not human, but it's human-like.
19:10Okay, you did answer the first part of my next question.
19:16So my next question is, were internal teams or passengers involved in testing real-time and real-condition laser?
19:25Market studies, collaborative choices regarding all the look and feel,
19:33and then great involvement from the passenger and from our team during all the data set construction,
19:43the data set building, and then testing, testing, testing.
19:46The technology is effective, but we wanted the technology to be...
19:53Well, we wanted the quality of the answers and the content to be as good as possible,
20:02and we managed to do that by involving our teams regarding what is the question,
20:07how do you answer, what is the turn of voice that, as Vinci Airport,
20:11we have to choose to create comfort and trust with the passenger.
20:16It can't sound obvious, but we have to do it collectively to make it successful.
20:22Okay, perfect.
20:23Now, also it is the same question, but I want to ask Christophe,
20:28what were the main challenges for Jumbo Mana that Lisek could work in a real place,
20:34in a real environment, with noisy places, so that it works perfectly?
20:40Yes, so the biggest challenge for us is what we call real world friction.
20:49Airport, like you said, is noisy, there are a lot of noise, children in background, etc.
20:58And for the other projects that we saw for public space,
21:03this project from competitors works fine in laboratory, in home,
21:10but when they put them in the field, it doesn't work.
21:15Because when we put a Jumbo Mana, for example an airport, it's not just about AI.
21:21We need to consider the environment and especially also the hardware.
21:26So at Jumbo Mana, we don't just need to put an AI to solve,
21:36but also all the system to work on this environment.
21:41So we benchmarked and we continue to benchmark the hardware, the microphone,
21:46the placement on the digital kiosk, etc. to also improve our model about multiple accents.
21:57Because they are a passenger in, like we said, we have English,
22:02but my French English is not the same like native English, etc.
22:08So we need to improve not just the solution and the AI, but also all the technology
22:16and also the hardware concerned by this environment.
22:20Okay, perfect. Now Cesar, I have a question for you.
22:24From the passenger's point of view, what value does Lisa add?
22:28Did you notice any impact on how people move through the airport
22:32or interact with this service?
22:35Well, we went live two months ago, so that's brand new.
22:41But the start is very positive.
22:43I mean, we are very proud of the project that we built together.
22:47The information, the data set is complex.
22:50When you're talking about an airport, you're talking about real-time information,
22:55web information, content that the airport staff can use on the ground,
22:59travel inspiration, so you have to open your data set to some travel sites
23:07about inspiration and help for travel.
23:12And I think we have a complete product so that we can see that
23:18passengers are very satisfied by the services.
23:22the interaction KPI are quite high, and Liza answers with a lot of relevancy
23:31because the data set is very rich, and we keep on improving and enriching the data set.
23:38One important thing as well is that the staff, once again, sorry to repeat, but that's important,
23:43is really at ease with the services.
23:47When Liza can answer, she answers.
23:49When she can't answer, she redirects to the staff on the ground.
23:54So it's a co-partner.
23:57And what we noticed from the beginning, the interaction keeps rising.
24:03The diversity of the question is increasing.
24:11And the quality of the question, with as well that feature of indoor mapping,
24:17way-finding, which is crucial in the product, is at the great level of quality
24:23regarding the customer feedback that we can have.
24:26Okay, perfect.
24:28Maybe we can see a small demonstration.
24:32And after that, if you want to try the Liza, you can go to the LRTP booth.
24:36You can find and talk to the Liza because the demonstration is a little bit long
24:40and we don't have a lot of time.
24:42But I'm going to show you how it works quickly.
24:46So here you can see the avatar.
24:50She did the introduction.
24:52And then we can choose the language, English or French.
24:58And then the person can ask any question.
25:00How to get to the nearest coffee shop, please?
25:03Here the question is to show you how to get to the nearest coffee shop.
25:08It means they will show a map with coffee shop.
25:16The nearest coffee shops are Starbucks and Masa and Company.
25:20Both are on this level.
25:21And you can reach them in about one minute.
25:26So if you are interested, you can go to the booth of LRTP to try it
25:30because we don't have a lot of time.
25:31But I'm going to ask Christophe one last question and then I'm going to conclude with César.
25:37So my last question is, what is your vision for Liza over the next few years?
25:42Any plans to expand beyond airports?
25:45Yes, so our vision is to create a true travel companion.
25:52Our goal at Jumbo Mana is to support the travelers and passengers from the first step through to the end
25:59of the stay.
26:00So our purpose is to go outside of the airport and to accompany the passengers to their destination and to
26:11the cities they are visited.
26:13And to make connections with all industries, from tourism, from mobility, to make sure when you are at the airport,
26:25Lyon Airport,
26:26and we have a flight to Paris to directly prepare your destination to see how I can get from when
26:37I arrive to Orsay Museum.
26:40And Liza can also talk about the traffic, the weather, the whole information beyond the Lyon Airport.
26:50And the mobility is a key feature for us, for the next development, to have this travel companion that can
27:00make sense with the passenger
27:01and to have information during all the journey.
27:06Okay, thank you.
27:07César, do you see Liza scaled or adapted across the airports in the Van Sea network?
27:15Well, when you were talking about expanding in the airport industry, I guess you were talking about the Vinci Airport
27:22network first and then other airports.
27:24But we see, well, we see customer innovation at the way, as you were mentioning, we've got the same vision
27:29in Vinci Airport,
27:30in that of being a travel companion or assisting our passenger all over the way in the travel experience from
27:39home, what we call from home to plane.
27:41And we've got digital assets that do the job for the moment, but we see that Liza can embody this
27:48one-to-one relationship, personalized relationship.
27:53And now we have kiosks on the airport.
27:57The idea is to demultiply all the kiosks and then to go directly in the customer digital asset.
28:04I mean, smartphone, mobile app and website.
28:09And regarding the whole network, the idea is to start with Lyon, but we can see that right now we've
28:16got a lot of other airports show interest in testing this solution.
28:22So if it's successful, I'm sure that we'll be able to roll it out on other airports in the Vinci
28:29Airport networks.
28:30Okay, thank you. Thank you for your attention, everyone. If you have any questions, we'll be glad to answer it.
28:36We will be here. And if you want to try Liza, you can find us at LRTP booth, so you
28:41can try the avatar and ask whatever question you want.
28:44Thank you very much.
28:46Thank you, ladies and gentlemen.
28:47Thank you very much.
Commentaires