 Hi everybody, good afternoon. Thanks for coming to my talk. Apology that I couldn't be there physically, but I'm still glad that you chose to spend your time listening to my talk. So just to make sure that you are in the right place. In this talk, I'll be covering a few things. The first thing is to explain AI Singapore. Who are we and what are we trying to do? The second thing is to share with you how we create open source solutions using the experience and knowledge we have by doing real life industrial projects. And the third thing, of course, is to share with you our available pre-built solutions and a quick overview of them. So hopefully by the end of this talk, some of you might find some inspirations or potential use cases for them. And if you have any questions regarding our open source solutions, feel free to let us know through our website or forum. We'd be more than happy to share a few more details about the open source solutions that we have built. So a bit about myself. I'm currently a principal AI consultant at AI Singapore. What I do on a daily basis is I help companies to embark on AI journeys. So it could be companies of any size, doesn't have to be SME or AMCs. So any corporations that wish to begin their AI journeys, be it defining a problem statements or building a simple AI model to solve a very specific needs. This is where I will come in with my colleagues. So that's my role at AI Singapore. At the same time, I've been second to the National AI Office of Singapore. So under the National AI Office of Singapore, I'm handling the healthcare portfolio. So I'm providing my AI technical expertise there. If you guys do some national healthcare projects. At the same time, also a visiting mentor at 500 startups. So this is where I provide some mentorships to the CTOs of startups. So in the evening, I'm pursuing my part-time master of computing at NUS. So you can see I'm quite a busy person. And in the past, I was from finance. So I used to be a training trader plus a Pokemon manager. So nice meeting everybody. And so be a copyright and let's get started. So what is AI Singapore or who are we and what are we trying to do? So my mission is to build deep national capabilities in AI and thereby creating social and economic impacts, grow local talent and build AI ecosystem in Singapore. So let me walk you through the overall statements from the left to the right through the graph. So we are funded by NRF National Research Foundation and we are hosted by NUS, right? And AI Singapore started in 2017 around June, July period. And what we do is that we have to coordinate scarce resources of AI because right now AI is very hard and talent series is very scarce. So we coordinate across different agencies and different universities and research institutions. So we have to coordinate efforts to do AI development activities gathered by real life needs. So what we do is that we go out to talk to the industries and find out what's the needs. And from there we do some coordination activities and help to build AI models for them. So at the same time, right? Due to the scarcity of AI talents, we also have our own in-house talent development program such as the AI apprenticeship program, AIP, AI for AI for everybody. This is a free three hours talk online which you can join with AI for AI, which is AI for industry. So we give you access to data cam for you to try out some of the coding exercises. And we have AI for students and AI for kids, right? So we go all the way down to the primary school level to train people in using AI. So under AI Singapore, we have three different pillars. We have the AI research, AI technology and AI innovations. So under AI research, what we do is that we invest resources in inventing a next generation AI techniques and algorithms. So this is where the fundamental research takes place, right? We do things like piracy, preserving AI, things like federated learning, things like that. So under AI technology, so this armed with soft national challenges. So I think that right now, one of the challenges you're trying to solve is on healthcare in tackling the three H, right? The hypertension, hypoglycemia and one other H. So AI innovation is where I am from. So AI innovation is small industry facing. So under AI innovations, we have three different teams. The first team is Learn AI. So under learning AI, we have various talent development programs that we have to teach people about AI. And I'll go deeper into AI AP later on, which is our fact sheet program. And there's also a program called Do AI. So under Do AI, this time we will do AI with the industry. So our 100 experiments and AI AP goes hand in hand. I will elaborate more on them later on. We have the AI bricks, which is the open source solutions that we have developed and we're trying to promote to the industry to use because they're free and open source. We have the AI clinic and AI discovery. So these two are more on the consulting arm. So this is where we help the companies to define a problem statements on AI, like out of the 100 problems that they have, which are the exact problems that they should focus on to solve. And we have data readiness as well. So one of the things that we notice is that in a lot of companies that we engage, they told us that they have data, but they are not ready for ML. So what we do is that we have the in-house program to help them do some data readiness check and do some data radius preparation so that once that is done, their data can be fed into whichever AM orders that they want to do. And lastly, we have the shared AI. So under shared AI, we have the forum, blogs and community. We are trying to build our ecosystem in Singapore where AI practitioners can share ideas with each other. So let me go in depth into the 100 experiments at AI apprenticeship, which I briefly described earlier. So under our 100 experiments, what we do is that we will work with any organization in Singapore of any size. As soon as they have problem statements and they have a data set, we can do something together. So under the 100 experiments, we have two tracks, the 100 experiments for research. So under the research research at the top part, the duration is typically 18 months. So this is where we will reach out to a network to get a professor to join the team to do the AI modelling together. So for a project to qualify under AI research, it must be something that is fundamentally challenging. It could be that the company needs a brand new algorithm to do certain things that can be done using out-of-the-box TensorFlow PyTorch modules. So they have to develop a brand new algorithm to solve specific business needs. So we also have this track called AI for Industry, or 100E for AI. So in this project, typically the duration is nine months and it's more translation in nature, meaning that, let me give you an example. Let's say if a bank come to us and they want to do a customer chain predictions. So typically these kind of problems could be solved using out-of-the-box solutions, right? SQL and PyTorch and so forth, you name it. So for this kind of problem statements, this is where we will work together over nine months period to do the AI models for them. And it is also under this 100E for AI, we will assign apprentices to the projects to work on this together. So this is where we match the demand and the supply together. So under the demand side, a company could have a demand for specific solutions to address their business needs. On the supply side, what we do is that we will have this AI apprenticeship training program where we require people to quit their job, to join our programs over nine months periods. So during these nine months, that apprentices usually two to three will be tacked onto the project. So along in that project team, there'll be like AI engineer, it could be me or my colleagues and a project managers. So these four to five men team will be assigned to this project, the work with the sponsor, project sponsor over the next nine months to solve the models, all the business problems that they are trying to address. So in overview, this is how the fact sheet programs of ours, the hundred experiments and AP works. So now I may be wondering, so what does this has to do with the open source solutions that we have, right? Through our engagement with different industry, we have worked on different sectors across different use cases. So let's go through a few of the projects I've done, which is pretty interesting. So on the top left, we have done a project with Sabara Jurong, to do a predictive lift maintenance, right? To predict which lift will break down in advance so that they can send attention style to do some predictive maintenance checks, right? We also done a project with agents, a company on credit scoring. So this is typically for the unbanked. So for example, if a mate come to you, we find a banking history, credit history, and want to borrow $50 from you, how can you access that percent risk, right? Without a proper credit score in place. So you have done that. So during my apprenticeship days, I have done a project with a kidney dialysis company to predict the hospitalization risk of a kidney patient, which is the project on our top right, right? So this is one of the projects I have done over the nine mile periods. So I was from the batch to apprenticeship program. I graduated, thankfully, Singapore likes me and I like them as well. So decided to join them as an engineer, subsequently becoming an AI consultant here. We also done a project on chronic room management. So this is chronic care. So what I have done is that they, we have developed this AI model to look at the wounds and to able to identify whether the wound is healing or not, right? So typically if you have a chronic wound, what you have to do is that every few weeks, you have to go to a polyclinic or any like doctors and nurses, right? They will open up the wound, the bandages and look through the wound and do some poking to see whether the wound is healing, right? So this is long, expensive and painful. So we have developed an AI model that is running on a mobile phone that can look at the wounds itself to assess whether the patient needs to go for further checkup or not. We have also our own NLP code switching on the right hand side, which is the speech-to-text, which I will do a quick demo later on. We also have done a abnormal detection with a security agency. So based on the movement of the vessel on radar, can detect whether this is normal or abnormal traffic. And I have done a project with a manufacturer as well, an electronic manufacturer to look at the, using committed visions to look at the equipment to see how likely that a customer is likely to return these things that is being produced. And lastly, we have done a project with Expedia. So Expedia has done a project with us on name-edited recognitions, right? Because like the way that people search in the text in Japan, right? There are many different scripts of Japanese scripts. So they all could refer to the same thing. So how do you improve the search results of the customers, right? So some of our projects that I mentioned earlier are announced in the news, right? That's why I can share their names. For most of the project that I've done, we have signed an NDA, non-discussion agreement for the companies. So for some of the projects, I can't share their company's names, right? If you do a quick search on Google, you'll see that a lot of our projects that we have done with big companies, right? It's all on the news, right? You can see that Expedia, AS, Singapore here, Cisabara, Juno, over here, Piotena, this is on cell therapies and on FHDB to do some construction safety. Using AI, right? So now I come to the main part of today's presentation, right? AS, Singapore Bricks. How do I link the hundred experiments that I talked about to the AS, Singapore Bricks? So let me explain using this graph here. So AS, Singapore is in a very unique position because we work across different industry of companies with different sizes. So we have a lot of engagement with the industry and sometimes if a project doesn't fit into the hundred experiments, because in hundred experiments, AS, Singapore will co-fund the project together with the company. So for projects under hundred experiments, it is to go through two stages of approval internally before they get the funding from AS, Singapore to co-build the model together, right? So for some other projects, right, which are more simpler in nature, we will take them on under our consulting arms and we'll do it. So usually for these kind of projects, for consulting projects, it's usually three to four months period. So we work with big projects, 80 months long under the hundred E for research, nine months long, a hundred E for industry and short term project under consulting. So we have projects ranging for super complex to quite manageable and easy to do. And at the same time, through all these project engagement, we spoke with the management of the companies, right? We got CEOs, the management directors on what the problems are facing. So what we then do is that from all the learnings and experience that we have, we take the common denominator out of them, not the data, not the proprietary algorithm, but our knowledge and experience. We pull them up together and we develop it into a bricks that can solve most use cases in the industry. So let me give you an example. So the first bricks are open source solutions. So let me clarify by bricks, we mean open source solutions, right? So bricks is just our acronyms, right? So the first brick that we have developed is for robotic process automation, right? Automating your keyboard and mouse to spare your productivity. So one of the bricks that we have is Tech UI. So Tech UI is open source RPA tool. You can see on the right hand side is a demonstration of how we can download the first racks, the FX rates from DBS and send it through email to whoever you want it to run. So this is one of our open source software. I can take a look at our website, which I will share with you later on if you wish to use this internally in-house or if you want to build a product out of it, please go ahead. The next interesting bricks that we have is called golden retriever. We name all our NLP tools, natural language processing tools under docs. So one of it is golden retriever. So this is like a FAQ retriever. So under this, what we do is that, right? In a typical chatbot, you need to prepare the data by listing the questions and the answer itself. So the unit use case of our golden retriever is that you don't need to specify which answer are tagged to which question. If you have a document of FAQ by loading it into the golden retriever, it can automatically pass out the semantic meaning of each paragraph. So whenever you type a question into the golden retriever, it can give you like the top three most relevant paragraph in that documents that you loaded inside. So the use case is, let's say for example, you are working as a salesman, your company has a lot of policies for the follow. So when a customer asks you, hey, can I have a payment extension? And you say, okay, let me check. You can type your questions right into the engine and it'll give you the relevant paragraph in the policy. The good thing about this is that it's not dependent on like control fine, right? The exact keywords match. It's dependent on semantic meaning. So let's say for example, light and love, even though they are not exactly the same, but symmetrically they are the same. So the engine will be able to recognize this, right? You can try out the demo on the website. Next, we have this another dog. It's called Korgi. So it's auto-labeler. So the use case is that it helps to accelerate your data labeling process, right? There's an issue here is that you might have, let's say you're in a call center or you are running a e-commerce store. You have a lot of consumer reviews, right? The issue is for you to train customer segmentation, customer sentiment analysis model, you need to manually tag each of the paragraph, right? So by using these tools, right, it uses a non-negative factorization behind it. You can read up more on the website, but what it does is that it helps to speed up the way you tag your data. And one more thing about NLP is that NLP is very complex, especially in Singapore context. Why is that so? Let's take a look at this sentence, right? They, we always mark one. So in these very short sentences, you have various languages combined together. You have Tamil, Mandarin, Cantonese, English, Hokkien, Malay, and English at one, right at the back. If you are to say this sentence into Google translate, it will give you, the transition result won't be that good to transcription because Google requires you to speak in that language that you select and in perfect English, right? But for most Singaporeans like me, we can't speak proper English, right? So we need our own speech-to-text engine that can handle our language. So this is where we have, we have this like, we have this NLP tools. Sorry, let me go back here. So we have this NLP tools that can handle different languages when you speak to it. So this is our own speech lab, right? So let me give you a quick demonstration because this is a very common, I'm not sure if you can hear it, but this, all the demonstration is available on my site. So let's hear it. Now, I hope the audio is available to you. So essentially what the guy is saying is that, he said, hey, hello. What I'm about to say means that I wanna make a police report in Chinese, right? So this guy is testing the engine by using both English and Chinese at the same time. And our engine is able to transcribe the languages in real time for two languages, right? So this is something that Google Translate is not able to do, right? We have a demo for our speech lab on our website. Please feel free to give it a try, right? So what is next, right? So at AI Singapore, we have, just recap again, we have the learning AI, share AI and do AI, right? So you can go to our website. Let me just give you a quick demonstration now on our website. So on our website here, you can go to AI Singapore Makerspace, this website. And under do AI, this is where all our bricks are. So for example, right, let me give you an example. So this is an example on golden retriever, which I mentioned to you earlier. So what we have done is that we have loaded in the COVID-19 database as well, right? So this is interesting. So right now, for example, if you can type in, what is coronavirus, right? We click fetch. So these are all the databases, FAQ that load up from the Ministry of Health in Singapore. So it gives you the top three results. So you can say, gonna buy us a large family of viruses, conic unicers, yeah. So this is what I meant by giving you the top three answers that you have. So let me give you a try. So is coronavirus dangerous? Okay, let me give you a try and you'll see whether it works or not, right? So the answer is, okay. So you can see one of the top two answer is many characteristics of the virus is still unclear. How it might affect people, severe disease, death in 23%. So the engine is able to roughly understand the symmetric meaning behind dangerous and give you back the answer the other way, right? So I have come to the end of my presentation. So let me just do a quick recap for you again. At AI Singapore, our mission is to help the companies in Singapore adopt AI and accelerate that AI journey, right? And what we are doing right now is that through these AI breaks or open source solutions, we are taking our industrial experience, combine it with the multiple projects that we have done. We distill our learnings and knowledge from doing this project and turn them into, find the common denominator and turn them into open source solutions which anybody can use. If you're interested in using our engine for going to receiver, all our source code is hosted on GitHub. So feel free to go use it to build products for the general good, especially nowadays with the coronavirus, there's a lot of false news, fake news spreading around. So it'd be good if there's a search engine which can have all the relevant authorized database answers that is within this engine that everybody can have access to. Won't that be great? So thanks again. Thanks for your time in listening to my presentation. I hope that you have learned something about AI Singapore and our machines. If you wish to work with us on our Henry E projects or to hire our apprentices who are graduating for our programs, feel free to go to just Google AI Singapore. It should be in the top few links. Come to our website, you can contact us and we can discuss whether, how it can work together. Thank you for attending my talk again. Stay safe, stay healthy, all the best. Thank you. Goodbye.