Intel Scaling at the Edge: Build-Your-Own Use Cases with AI Building Blocks
Previously featured on Intel® Scaling at the Edge video series.
- Joy Chua - EVP of Strategy & Development at meldCX
[00:00:00] Introduction: Welcome to the Intel Scaling at the Edge video series. In this video, we will learn how meldCX is creating more connected retail stores, safer workplaces and smart healthcare spaces by enabling machines to see the physical world with speed and precision. Joining us is Joy Chua, the executive vice president of strategy and development at meldCX. Joy, take it away.
[00:00:27] Joy Chua: Thank you Intel for the opportunity and thank you everyone for your time. Again today, I'm really excited to be able to take you guys through meldCX, which is our company, as well as our AI platform Viana, and what we can do as a business to help you make AI and automation a reality, the latest advancements in artificial intelligence.
[00:00:48] Machine vision and the internet of things, allowing organizations to make the most of anonymized data collected from cameras and census. The aim is really not just to passively monitor environments, but also actively gather insight, make smart decisions and automated processes so that we can help businesses.
[00:01:10] And that's what we essentially do. meldCX stands for ‘melding customer experience’. And what we do as a company is specialized in using AI and intelligent edge technologies to provide better customer experiences. And we've been on the journey for a while now, and we've realized. You know, in this journey, a lot of customers often see gaps and challenge us in this edge and AI adoption process.
[00:01:36] Some the questions that we usually hear are things like, you know, how do I scale? How do I grow with the resources I have? But still make sure I'm in line with security and compliance. AI is not an unfamiliar word to a lot of us, but it’s still pretty hard to palate and digest for a lot of organizations. We find people have a lot of plans, but turning those plans into reality sometimes still requires quite a big leap.
[00:02:04] And so that's what we've created. We created Viana, which is essentially an insulin platform that allows organizations to make your AI and automation a reality based on what you actually need without being forced to invest too much in data engineering and ML resources.
[00:02:22] And so a little bit more it's helped about, you know, security, privacy and compliance. And on this slide, you'll see the top six reasons why we're trusted by some of the world's biggest brands. And I'll do a little bit of a deep dive as well into our edge-base anonymized approach. So some of the key reasons that, you know, help us really stay within those ethical safety and compliance boundaries are
[00:02:48] One, we make sure that all faces are blurred in the streams and processed at the edge. That means that when you look in the Viana platform, you don't actually see someone's face. So all faces are blurred. Like you can see on the left of my screen, with that particular lady wearing a blue blazer, we actually make sure that faces are pixelated, which means that no personal information actually passes through our Viana platform.
[00:03:16] Second, we make sure that every person is turned into an anonymous token. So instead of being known as Joy, for example, which is myself, I will be known on the system, like you can see on the left, as token with the numbers ending 002. So that's who I am. And so there's no personble identifiable information that’s stored.
[00:03:37] The second as well is on the Viana platform, we make sure that we don't store or stream identifiable sensor data. So what I mean by that is in the case of a loss prevention scenario, for example, um, we're able to monitor the anomaly, process that anomaly and have the anomaly sent to the right person. Say, if we detect token ID 001 has been standing in front of the shaver shelf, for example, in the pharmaceutical section, um, he's picked a couple of things off the razorblade sites.
[00:04:14] He's looking over his shoulder multiple times. We flag that as an anomaly, we send that off to say security staff and personnel. When the security staff and personnel receive that notice, all it provides to them as really a link with, you know, the timestamp of the video clip where the anomaly has been detected.
[00:04:32] We don't actually point them to our platform. We just notify them of where to look at that particular snippet or segment of video on the CCTV system. And so that means that, you know, we don't actually show. That particular video clip on the Viana platform. And that's how we ensure that privacy is protected that way.
[00:04:52] But the thing that we do that I want to spend a bit of time focusing on is we also realize that a target persona is really more than just your face. And so in terms of, we don't really do things like facial recognition, but what we do is we add more depth to that persona by combining things like object, as well as non face behavior.
[00:05:15] So, what I mean by that is on top of some of the usual stack that we can gather, or the person, say using myself as an example, um, female age within in this range. I also pick up object recognition, such that, you know, um, Joy has short black hair. She's also wearing glasses per se. So, what that means is I'm able to add that much breath to that target persona, without identifying that person, but get enough that to extract it, to be able to provide information that might be useful for this persona.
[00:05:46] So say if I was walking on a supermarket show, for example, and say, spectacle wipes were on sale. What that would be is as I was walking you through a digital signage, could then flash out with things that were suitable for me and people who are wearing glasses that were on sale up on the screen, just to be able to send more useful information to me, that way.
[00:06:09] The other thing that we also do is non-face behavior. So, I use loss prevention as an example of non-face behavior that we put track and monitor as anomalies. But I also went to the top about using non-face behavior for metrics purposes.
[00:06:24] And for metrics purposes, what I meant was, you know, instead of tracking anomalies, you can also track things like: how is a particular target persona interacting with say, A Macbook on a bench. Are they looking at a 15 inch versus looking at, you know, a 10 inch tablet. How are the hands actually behaving? How are they interacting with things on the bench and providing those kinds of metrics data back to the business, to be able to understand, you know, this target persona, say for example, my age group, do they prefer tablets over laptops? And whether we should stock more of this product. So these are some of the ways that we make sure that we keep anonymous. We stay within the confines of privacy, as well as, security and compliance.
[00:07:14] So I spoke a lot about how we look and capture anonymous data. Other things that we do in how our customers engage with us is really, you know, in three ways. And we've cultivated AI in this kind of three modular ways, because we know that the AI journey sometimes gets a bit intense, or it's such a big word to actually grapple and digest. And so we've got three key ways that we engage with our partners so that no matter where you are at on your data or your AI journey, we always have the right engagement module for you. And so if you see on screen, we have three engagement models.
[00:07:52] So the first is Start, the second is Grow and the thought is Evolve. And essentially Start is our term for businesses who are just starting their AI journey. So in this particular scenario, we'll come to you with our ready to go or off the shelf modules. Which means that you're literally able to go into our Viana platform, download the modules that you have, deploy them on your sensors and really be ready to go. So that means that if you want to just do a quick check on, you know, people in the space, they'll get people counting modules, that's all available for you. You can quickly download it and work together with your team to deploy that and get analytics quickly to prove your business case.
[00:08:41] The second is our Grow engagement module. And our Grow engagement module is perfect for customers who have dabbled in that AI engagement space before. They know a little bit about AI. These customer groups usually already have some form of existing AI/ML modules, but are looking to add more. So they've dabbled in that and they realize that this particular model or algorithm that they have doesn't necessarily cover off all their business requirements. And they're looking for a supplement that can have ready to go modules as well as use what they already have. And so that's our Growth engagement module.
[00:09:18] Our last engagement module is our Evolve engagement module. And this is for businesses who are very clear on your business requirements and are not all about proving a business case because they have a hypothesis statement or they have a very clear requirement's need.
[00:09:35] And so in this use case, we usually work with customers to train new AI data sets. It's all about personalized use cases specific for that customer. And what we find is that even in this group, even though there might be really specific use cases and we work with them to train specific AI data sets, our range of modules are usually able to work across these sets as well.
[00:10:00] So that means two things. One, you can reuse some of our ready to go modules. Two, you know, you can reuse some of our dashboards that we've already created for customers. And three, that also means that you don't have to invest so much kind of, ML and data engineering resources to do that. So you'll be able to stand up a pretty quick personalized use case by yourself as well.
[00:10:22] So I've started to show a couple of examples of what I mean by ready to go and how, you know, some of the modules that we have available and how that could tie in with our three growth engagement modules. So if you see on screen here, you'll see we're using safer workspace and we're really a corporate enterprise verticalas an example. I'm going to showcase different verticals, how we have ready to go modules that can split across all three and how we can then work to create personalized use cases for you guys as well.
[00:11:04] So the first is our corporate workspace vertical, where you can see on screen everything in that little white box is everything that's ready off the shelf and ready to go, basically. So we have a couple of modules, as you can see, from something as standard as entry monitoring, where we're actually tracking foot traffic in and out, the number of people in a particular entryway, for example, to more complex off the shelf modules which includes our SAMi module, which is our cleaning and compliance module.
[00:11:27] So in this module, we're combining a sensor feed and overlaying that with heat mapping technology. So what that means is in a high traffic environment, like a cafeteria or boardroom, for example, we’re able to track the amount of touches, the amount of people who are sitting there at the moment and set things like contact and time thresholds.
[00:11:51] So instead of having a really standard, you know, “I'll clean at 9, I’ll clean at 12 and I'll clean at 5:00 PM”, we're able to then set those thresholds. And when those thresholds are met, we'll be able to deploy cleaning or operating staff to clean and turn over that area. And so that's all about, you know, and really, I think, good in this time of the pandemic where we're able to do that and ensure that we're keeping a safe workspace for everyone.
[00:12:19] Other things that are also available, ready to go, also include things like our audience measurement and content effectiveness modules, where it's all about measuring target personas interaction with digital content on screen. So things like age, gender, and sentiments are captured. And with content effectiveness, we're also able to understand if the target persona has watched an ad through.
[00:12:44] So by that, what I mean is we're able to do eyeball tracking to see if someone has actually watched a 15 second ad from start to finish, for example. And if they're looking at this particular spot on the presentation, so all of that provides really valuable feedback for product marketing, retail marketing teams, to be able to then create that specific, um, ad to the right target persona.
[00:13:10] The last thing that's also available off the shelf is out zone engagement module. In zone engagement, we're able to basically track and stitch, you know, anonymous personas as you go through the different zones in a particular environment. So we'll be able to see if someone has spent more time, say in the boardroom and specific areas of, you know, a retail lab, for example, and who's the top persona that's visiting these areas.
[00:13:39] So, this is an example of some of our ready to go modules. And I'm going to show you the next slide, which as you can see, we've converted some of those modules and applied them to a retail store. So our modules are essentially what we call horizontal modules that can just be taken and applied in different verticals for the different use cases.
[00:14:00] So you can see on screen here, anything that's in the white box, are some of the ready to go modules that I've introduced in my previous slide and applied in the retail scenario. So you'll see really familiar modules, like I mentioned before, about audience measurement, where we're tracking, you know, personas and how they're looking at digital content or digital signage content.
[00:14:23] You can also see similar modules, like foot traffic counting, or rather entry monitoring. And you'll see our cleaning compliance module on there as well. And this scenario, um, we've also put -- this is our Growth module. If you recall, we've also put an add your own module there. So in a lot of retail customers, what we've also found is in this scenario, a lot of retailers usually have already dabbled in some form of AI before.
[00:14:54] And what they do in the AI space is they already have some models and algorithms. And so what we find is that's really a lot to do around product. So they usually have some form of AI algorithm around product. And so we can easily take that AI module, ingest that into our Viana platform and then be able to overlay that with some of our other technology, which you can see on the screen here in that red box.
[00:15:20] It's all about making sure that we can overlay product with, you know, things like how many people are attaching this particular Samsung phone, for example, who is the target persona that's interacting most of it, which is a combination of, you know, product algorithm together with our zone engagement module, to be able to produce our shelf engage my module, as you can see up here. So these are all, some really valuable statistics or analytics, data that we can really take, extract and surface into dashboards that can provide information back to the business on how customers are interacting with their products, what they should invest in and who their target personas are as well.
[00:16:03] Last but not least, I've also taken, you know, our ready to go modules and added that to some of our personalized modules as well. So in this case we’ve lifted some of our core modules, which are still in the little white box, and apply that in the healthcare space, for example. So you'll see really familiar modules like our entry monitoring module that's on there. You'll see our cleaning compliance module and audience measurement modules there, as well as our zone engagement modules, which are all ready to go modules.
[00:16:34] In this scenario with healthcare, which is an example of a vertical that usually has really specific, custom use cases that they want to achieve. You'll see, on screen here, the little black box of the exclamation mark, and that is some of our personalized modules that we are creating for some of our healthcare customers. So some of these include things like slip-and-fall; so making sure that, especially in an aged care facility per se, if a particular patient has fallen, being able to deploy, you know, nurses and idle care workers straight to the scene so they can monitor and help the patien, to things like aggressive behavior alert, where something has a potential to escalate, say in the event of drug use or alcohol abuse, that we can deploy, not only health staff, but also security staff to the area to help, you know, de-escalate the situation.
[00:17:33] So these are examples of modules that are heavily based on environmental factors and perhaps, you know, target, um, persona groups as well. And things that we can work with your business to be able to personalize and then deploy. And in this scenario, you can see that often it's really a mixture of hybrid approach where we're overlaying some ready to go modules that are off the shelf, together with custom use cases, to be able to have AI make a really distinct impact in your organization and your business, and prove business cases really quickly.
[00:18:10] And, you know, last but not least, little shout out to Intel and some of the technology that we use to bring it all together. We all operate on edge devices like Intel processes. And we use that to be able to, you know, process things on the edge, even more quickly. And we also use OpenVINO to be able to optimize all our AI modules. And as you can see here, and that's how we form, you know, our AI platform bringing the best of cloud and edge together.
[00:18:40] Also, I spoke a lot about verticals and, you know, we just wanted to do a quick call out into some of our target verticals and some of the verticals that we've seen, you know, immense success and have multiple use cases in, um, you can see on screen here, we've got retail, hospitality, industrial healthcare, you know, smart cities and workspaces as well as governance.
[00:18:59] And we've put a little bit of some of the key modules that we found had the most traction in these verticals. So for retail and hospitality, for example, some things that our customers really like and have continuously used are our product and zone engagement modules, as well as things around theft detection and loss prevention, or things around loyalty as well, has been very popular with our retail customers.
[00:19:24] And in our industrial customers we see a lot of need in AI scenarios around warehousing and back of house efficiencies. And things like, you know, I mentioned before in healthcare, which we spoke about, a lot of it is around, you know, slip and fall detection and all about enhancing patient care.
[00:19:41] So, yeah. Thank you again for, you know, your time today. If you'd like more information, please don't hesitate to visit our website or you can scan a QR code on the screen to download our Intel whitepaper today. Thank you again.