This is a recap of meldCX's session "Accelerating the IoT Journey with AI Building Blocks" at Intel AI Summit 2021.

‍

Life on the edge! A Gartner study reveals that 85% of AI projects won’t deliver. Businesses around the world have realized the benefits of AI, but not many have the tools and knowledge to implement such complex solutions.

In this session, meldCX shares our vision in democratizing AI — making it consumable for all industry verticals, and how we've achieved that by building an AI solution that acts as building blocks. Discover how we've deployed AI solutions on Intel technology for the likes of Australia Post and Westpac, the challenges they faced, and how they overcame them. Explore success stories around handwriting recognition, customer behavior measurement, digital signage content effectiveness, and more.

Featuring:

• Joy Chua - EVP of Strategy & Development at meldCX

Transcript:

Joy Chua  0:05  
Hi, everyone, thank you so much for your time today. My name is Joy, and I'm part of the meldCX team, leading strategy and development globally for the business. Today, I would like to share about our team, and how we have created AI building blocks using our analytics platform, which is called Viana, to help accelerate the IoT and data driven journey for our customers and partners. So a little bit about meldCX and who we are.

Joy Chua  0:38  
So with meldCX, meldCX actually stands for melding the customer experience, and who we are as a team is really all about driving premium customer experiences using edge and AI technologies. And so we came up from a broader group of companies that has, you know, backgrounds in hardware that we've been given the mandate to just take best of breed technology, combine that with best of breed hardware into the software platform that we are now. And so we work with partners like Intel, Cisco Meraki, Microsoft and Google. And using our AI platform Viana. Essentially what we do is give autonomous devices, the ability to see analyse, engage, and make actionable decisions seamlessly.

Joy Chua  1:30  
So what we really like to say is that we teach machines to see like humans. And use our portal to surface and consolidate data to tell a story, create an intervention, or drive action or insight.

Joy Chua  1:49  
So a little bit more about our portal. Viana stands for vision analytics. And what we have is a simple to use end to end platform, build using best of breed technology, from Intel, Google and Microsoft that can identify humans and objects, monitor anomalies, to present logical and unbiased outcomes.

Joy Chua  2:14  
We have rolled the product out in three continents, and have customers in multiple verticals like retail banking enterprise. And what I really resonated with his customers is the way we approach AI. So I think AI in general is a really big beast. But what we really seek to do is to democratise it. So we want to break it down, make it easy to consume and be palatable. And so I'll talk about this more, and the building blocks we have later on in the presentation. But what we found really resonates with these customers as well is our age base, anonymize approach. So on the screen here, you'll see six key things that we adhere to as a brand. So from a privacy perspective, we ensure that all faces are blurred in strings, we capture from sensors, you know, like a Meraki camera, and that no identifiable data is stored or streamed, which is really important, because we want to make sure that we process everything at the edge.

Joy Chua  3:18  
From a security perspective, our platform is also used for payments. And we've achieved what we call PCI level one certification. On top of that, because of some of the clients that we've worked with, you know, we've undergone multiple security and risk assessments, especially with clients in the logistics and banking space. And lastly, as a vision analytics company, we have also learned that a person is more than a face. So as a team, we track spatial coordinates. So this could range from things like age, gender sentiment, but one important thing to note is we don't do facial recognition. But what we do is add more facial coordinates like the ones we spoke about before. And we add more detail and depth to the anonymized persona by combining things like object. So you know, clothes a person is wearing, and non face behaviour, like gait and aggression.

Joy Chua  4:15  
So if you see on this slide here, this is an example of all the points of reference that we can track on an anonymized individual, which is then tokenized and can be slotted into a unique persona to provide further insight.

Joy Chua  4:31  
So one of the other key things that the Viana product does is something in what we call re-identification. So what we do is aggregate video feeds across multiple cameras, using physical characteristics like apparel, and accessories to track the anonymous person on their journey through multiple store, multiple zones, apologies, in per say a retail store.

Joy Chua  5:00  
And what this does is allow us to do the following. So it allows us to understand how they're interacting across different zones near the time they're standing at each stone, which really gives us a good indication of products or items that they may be interested in.

Joy Chua  5:19  
So how do our customers and partners engage with meldCX? And so as I mentioned earlier in the presentation, you know, we spoke about developing building block, and our team's goal to democratise AI, and make it consumable for all verticals. So in general, there are three ways we engage with customers.

Joy Chua  5:42  
So the first way, you know, is off the shelf modules that we have are what we like to say ready to go. And the second is, you know, to add your own AI model. And the third way is where we build personalized data sets for Greenfield use cases of our customers. And we built these three building blocks and provided multiple ways to engage with the meldCX team, so that we will be able to service organisations who are in different stages of their data and AI journey.

Joy Chua  6:14  
And so how we see this is, if you're starting out, and you want to quickly validate a use case for business, you know, pick one of our off the shelf, we're ready to go modules, it's designed so that you can configure without needing a developer or a data engineer, which makes it really easy and simple. And if, you know, you started, you know, you have modules, or models that you've created, but are really looking to expand your use case, you know, we've got another way to engage with us, which is all about adding your existing AI and ML models and algorithms on our platform.

Joy Chua  6:50  
And the third, obviously, if you know, there's something entirely Greenfield that you'd like to try, you know, we can work with your team to train new AI data sets, and support new use cases. And so I'll unpack a little bit about how, you know, customers have been in the various verticals have been making use of this building blocks in the next few slides that are coming up.

Joy Chua  7:17  
So as you can see on the screen of this particular slide, this is an example of all our available off the shelf modules applied in a connected workspace scenario. So some of these modules include entry monitoring, which is you know, a module which we used to track, you know, the ins and outs of foot traffic to see how many people are in a zone, who's coming in and who's coming out. We also have in, you know, ready to go modules semi, which is basically our surface tracking, cleaning and compliance module, where what we do is we overlay heat mapping technology, with camera feeds to monitor high traffic or high touch environments to understand if we need to dispatch more cleaning stuff to the area, instead of waiting for set cleaning times. And what this particular module does as well is it provides a digital cleaning manifest, which means that you're able to track who has cleaned it, at what time and which particular zone their claim. So imagine instead of having you know, your clipboard with tick box and crosses, and you'll be able to have something electronic this time, so digital manifests.

Joy Chua  8:29  
Lastly, as well, one of our ready to go modules include zone engagement, which I mentioned before, is what we do with tracking anonymized personas and movements across the different sections in the floor. And so all these modules are available now, and use the same hardware infrastructure to enable, which really means that our customers and partners can use, you know, the same upfront OPEX costs to prove and validate use cases before scaling. And so you're able to really try start and understand your data driven journey, before having to make you know that that huge investment when it comes to productization.

Joy Chua  9:10  
This next slide shows all the existing ready to go off the shelf modules, which I spoke about just now, but applied in a retail scenario. So if you have a look, it has one exception, however, in the little red box that we've boxed up, which is you know, the Agile or model option. And in this scenario, a retail customer could bring existing models for their retail product and add them to our platform. And what this really means is as node, we could use a customer's existing data sets to identify a product. Or we could just start here by using the standard meldCX Viana model and pinpointing the exact location in the shell so that you don't have to train more data.

Joy Chua  10:02  
And lastly, in this scenario for healthcare, as you can see, we are continuing to use the ready to go modules like entry monitoring, zone engagement, and our surface when it's modules. But if you have a look, there are some personalized modules like for detection and aggressive behaviours, which we've mapped out in the little black boxes with the red exclamation points. So these are our personalized modules, which means that there's usually additional training that we do as a team within a particular environment, like a hospital hallway, so that we can really personalize this for your customer for the partner.

Joy Chua  10:47  
And so I guess the question is really, you know, what Intel products do they use? You know, rather, what have we leveraged from the Intel ecosystem to deliver our AI solutions. And so, as a quick summary, you know, we use two key products from Intel. If you recall, early on in my presentation, and I mentioned that we build and process or models at the edge. And we use Intel Core processors ranging from you know, i5 to i7, and the select edge devices to give us that additional grant that we need for analytics. On top of that, we also use open vino to optimise our models. And so to illustrate, we have a little video here to show you how OpenVINO powers Viana deep learning.

Video playing  11:37  
Machine learning models constantly infer. But when there's a high demand to infer models could run low on energy and unable to keep up. We can power up the model, making them process heavier loads of data and perform faster and more efficiently.

Joy Chua  11:55  
So where are we rolled out? And who do we service? You know, we'll take the next few slides to show you some of our customers and the exciting use cases that we've worked with, together with them commenting on our product to market.

Joy Chua  12:09  
One of our customers is one of the big four banks in Australia, Westpac, Westpac uses some of our ready to go modules like audience measurement, and content effectiveness, to understand how the inbound customers are interacting with digital content in branches. Through this, they are able to understand, you know, what really counts as effective content, and which audience segments are more receptive to the various messaging that they have in store. And so this slide that comes out there really, you know, illustrates how we combine Viana or what we've captured from vision analytics and the data we've got, we've combined that, together with proven play data. And what we essentially do is we detect, you know, facial coordinates that are used to inform things like age, gender, and sentiment. So are they happy? Or are they neutral, as well as gaze to see if you're watching the content or his entire duration of play. So these stats are then compiled and delivered in monthly or quarterly reports back to the banks, retail and product teams for analysis. And this is where they can really deep dive and understand, you know, is this particular piece of content, like a first home home, first home loan, tailored to what you're a younger demographic, or should I be targeting more of a mid range? Know who's really interested to hear this in bank.

Joy Chua  13:39  
For those who are interested as well, we also have a whitepaper on this use case. So please visit our website or our LinkedIn page, to download the white paper and to learn more.

Joy Chua  13:53  
We also have a hybrid approach with AI data with Australia Post. So the image you see on screen here is an example of a personalized AI use case that we've worked on with the Australian post team. And this particular use case was also demonstrated together with Intel at NRF, which is the National Retail Federation show in New York, in 2020. So what this essentially was, is, we worked with Australia Post to come up with a custom kiosk, that really was all around automating the pathologic process. So we were you know, some key outcomes included, you know, reducing person to person contact and queue time. So, imagine if you could put a parcel on the scale, you know, have your handwriting for addresses be recognised, you know, that kiosk was also where your parcel, you could pay for this and then drop it off at one of the, you know, parcel shoots. So, that's essentially what the use case was. And is this an example of a personalized use case that, you know, totally Greenfield example that we've worked with Australia Post on. And so, you know, this is part of some other projects in the pipeline, you know, that combines ready to go and personalized parcel logic and behaviour tracking with the team at Australia Post. So all I can say is, you know, continue to watch this space with what, a lot more than it's happening of this particular customer.

Joy Chua  15:24  
And in this next slide, we're also working with a large electronics retailer in Australia, to really use vision analytics to track customer flow and behaviour, not only inside the retail stores, but also outside. And, you know, some of the core outcomes they wanted were to use vision analytics to understand, you know, which zones, customers are spending the most time in. Number two, you know, what products they interact with the most frequently. And then number three, because we do have some cameras pointing outside of the store, to also understand how much of the percentage of customers are actually, you know, more inclined to to click and collect instead of going into store.

Joy Chua  16:06  
And I guess that - lastly, you know, really, how can we help you be successful in implementing your AI use case? So, you know, as a business, I think the one thing that we really believe in, is to start small, prove value, and then scale. So if you're just starting the journey, you know, as I mentioned before, we have a range of ready to use building blocks that you can take advantage of, for a quick POC, before you scale. We've got modules that range from, you know, things like vehicle counting, to audience measurement, content effectiveness, you know, even entry monitoring, so these are all ready to use, you know, you can deploy them quickly and easily. And if you have a greenfield use case that you'd like to explore, you know, please reach out to our team, and we'd love to hear from you. And one thing that we usually do is, we love to work with our customers and partners, on what we call personalized solutions, jam sessions. And it's in this sessions that, you know, we sit, we brainstorm what you are facing in your industry, you know, what you're facing, and what does hold the key pain points, you know, partners like yourselves or customers are facing, and find a joint PA to really help you in your data, and your AI journey. And on top of that, we also have, you know, heaps of content or collateral to assist. So, you know, please do visit our website, you know, our LinkedIn page for white papers, blogs, are just to really get in touch.

Joy Chua  17:49  
And so, you know, really want to thank everyone again for your time today. And a huge shout out to the entire team, you know, for the opportunity to present. Thank you so much. If you would love to connect. I put my details in the link on the screen as well. So we would love to hear from you. So yeah, please reach out and you know, we're here to help you plan your data driven journey together.