Microsoft Community Insights Podcast
Welcome to the Microsoft Community Insights Podcast, where we explore the world of Microsoft Technologies. Interview experts in the field to share insights, stories, and experiences in the cloud.
if you would like to watch the video version you can watch it on YouTube below
https://youtube.com/playlist?list=PLHohm6w4Gzi6KH8FqhIaUN-dbqAPT2wCX&si=BFaJa4LuAsPa2bfH
Hope you enjoy it
Microsoft Community Insights Podcast
Episode 34 - AI Innovation with Azure from Concept to Creation with Harshavardhan Bajoria
Harshavardhan shares his journey as an Associate Product Manager working with Azure AI solutions, from development challenges to community engagement. He discusses creating AI-powered interview systems and adaptive mock tests that serve millions of users while explaining how combining Azure and Gemini enables more robust solutions than using either platform alone.
• Seven-time Microsoft certified professional working with AI solutions at Unstop
• Developed an AI interview system that responds like a human and challenges vague answers
• Created adaptive mock tests that generate personalized learning roadmaps for users
• Combined Azure AI services with Google's Gemini to leverage strengths of both platforms
• Faced challenges with documentation and support when working with preview services
• Recommends Azure's prompt shielding feature for securing AI implementations
• Hosted 55 community events teaching AI implementation in the past year
• Focuses on helping students understand when and how to implement AI solutions
Hello, welcome to Microsoft Community Insights Podcast, where we share insights from community experts as they have fame in Microsoft. I am Niklas. I'll be your host today. In this podcast. We'll dive into AI innovation, with Azure concept to creation. But before we get started, I want to remind you to follow us on social media so you never miss an episode, and it will help us reach more amazing people like yourself. Today we have a special guest called Arvindan Parag. Sorry, I pronounced it wrong. Can you please introduce yourself?
Speaker 2:Yes, so hi everyone, hi Nicholas. So I'm Harsh Vardidyan and I'm currently an associate product manager at Unstaff Seven times Microsoft certified, google cloud digital certified, github certified and currently a GitHub campus expert also, and I just do creations out there, learn about AI and stuff and just contribute out to the community by engaging with different people, just like I am currently doing with Nicholas out here.
Speaker 1:Okay, nice. Could you please tell us a bit more about yourself? So what does your job involve as a product manager?
Speaker 2:So in my job currently, I look into the AI vertical at Unstuff, where my job is to actually find out how can we integrate AI services, and it's mostly currently Azure AI services or Gemini that we are currently using. So how can we develop using it? How can we integrate the new stuff that's out there? How can we build something upon that can help impact different users out there and, especially since we have a very large community of around 22 million users, how can we engage them daily so that they can grow and get the dream job that they're looking for?
Speaker 1:okay, so it's basically. Is it basically like a ai research? So just be sure, any new services, innovation in azure ai, whether it's a foundry service, whether it's a different part of like, whether it's semantic kernel or so so it's both uh, so I'm doing like the search also, and then the development part also.
Speaker 2:So I have a team with me and um, so I with them. I explore things with them and, after exploring creating the POC, finally developed the whole logic with them so that it can be developed and made production level ready so that it can be used by millions out there.
Speaker 1:Okay, so let's talk about some of the POC that you use Azure AI for. Can you give us, not saying client names, but any use cases that you use in your project?
Speaker 2:Sure, so like we have been using Azure AI services I can't name specifically where but like I can tell you in general about the projects of that work, yeah, scenario.
Speaker 1:You can say about scenario and what it is and the impact on it.
Speaker 2:Cool.
Speaker 2:So, like projects like I'm currently building a project where it's about detecting which person is speaking whom.
Speaker 2:So suppose, like I want to check whether that person is good with translation or not, whether I want to check that, okay, that person is asked to read a sentence or talk about a topic, whether that person is good with grammar, good with vocabulary, good with other XYZ stuff.
Speaker 2:So what I do is maybe use Azure Pronunciation Assessment and specifically explore how things are working in the backend and then on top of it, different models like we have model catalog present in Azure AI Foundry. So, on top of it, adding an LLM chain so that we can explore it and get much more detailed reporting out there that can be used. So that's a personal project that I'm building upon and, similarly, a personal project is like a complete pool library which you can can say. So if someone wants to go ahead and prepare for some of the exams out there or just go ahead and prepare for a hackathon, then maybe how can they go ahead explore different stuff out there using gpt 4o mini so again, it's present in model catalog out there. How can they go ahead, build upon that? And, uh, it also uses the sentiment analysis as your services, which you get, so understanding the sentiment of the user so that it can modify the behavior accordingly okay, uh, what are some of the?
Speaker 1:did you create these projects as it just has in your own time, or is it something with all in you in your workplace you had to do for poc?
Speaker 2:so, uh, the projects that I mentioned are being currently uh built during my own free time, because I like to explore things. Um, I can't, uh like, to be very honest, tell about the things that I'm doing at Unstop right now, like, what are the things that we're using, since I need to get permission for that.
Speaker 1:Yeah, but I think what we want to concern about is the scenario that you use at work. So you don't have to be suspect, but you can just give any scenario that you think has impact on people, whether it's a client or stuff, but you didn't have to go into detail okay, cool.
Speaker 2:So it's currently uh, I have launched a new thing which is called ai interview. So it's like, uh, the normal ai interview, which you might have seen in different places. But the thing that makes it different is it is like a human. It will respond to you like a human and the best part is, if you give a very general or vague statement let's suppose you just say that, okay, I am a very good person at this particular skill or I had a good chance of learning that skill it will specifically ask you the reason.
Speaker 2:So that will set up, set it apart and also like a coding platform integrated where you can have coding interviews, you can have email writing interviews, you can have different structural interviews out there. So that's the thing I've launched recently, along with, like, ai forward mock test. So if you want something like a continuous mock test that changes according to your needs and give you a strength, weakness and a roadmap complete roadmap of how you can actually learn that skill where you are lacking in, it, gives you all of it and it's using AI, again in the backend. Again, it's a combination of Azure and Gemini, so it's working together hand in hand so that we get the most of both the worlds out there. It's working together hand in hand so that we get the most of both the worlds out there.
Speaker 1:Okay, did those two use case scenario? Did it ever reach production? Or they're just POC, just for proof of concept?
Speaker 2:They are production level right now and they are being used, so it's their live on the website. Both of them and the first one, ai. Ai interview is currently in progress, while mock tests are being used by students out there. Our users, 22 million around.
Speaker 1:Okay, how are you using? What tools are you using to develop these scenarios? Are you using GitHub Codespace? Are you just using Azure AI? Are you just GitHub Codespace or are you?
Speaker 2:just using Azure AI or are you just using Python? So currently, based upon our tech stack, it's Angular, but we are using a combination. So it's like we are again choosing the best of the things. So we are choosing Angular, react, laravel and both of it, making it a combination, and then for the AI services, as I mentioned, it's mostly Azure plus Gemini combined together or some of the AWS services. So it's like we are taking the best of the cloud computing service that we can get and combining them together to create a robust chain out there so that we can create something that is accessible out there, that is helping the people get the best product that they need.
Speaker 1:Okay, so let's just start from the basic scenario. Did you have to write Angular code in VS Code and then use any CIC, divided Azure DevOps or GitHub Actions or anything, or you just have to just deploy it to Azure?
Speaker 2:So for that particular thing I was not responsible because I am a product manager, so I don't look into the pipeline stuff of it and the development stuff of it. So I don't look into the pipeline stuff of it and the development stuff of it, but I believe we use automated CIC deployments by GitLab. Not sure about it, though.
Speaker 1:Okay, what are some of the challenges that you had to face when you overcome these AI projects, when you had to do them between your teams or POC?
Speaker 2:So the major challenge that I have interacted with other people out there they find is actually getting the resource of it. So like when you want to learn about a new service, for example, let's say GPT-40-0 mini-reality, which has been run recently but it's under preview mode, so the resource isn't available readily out there. Even if you go to Microsoft Learn or Microsoft Docs, you can't find the documentation made available right there. The second thing is getting the support. So suppose you are building a particular project and you are not able to get the library for it. So if you want a library in React or a library in Angular, since it's in preview, it becomes a bit difficult in the current scenario to get the libraries and getting to understand how the LLM is, understand how the llm is working in the backend, because, again, it's in preview, so the documentation isn't available ready. So yeah, these are the two main challenges which people face and in fact I also face during the development of the air projects out there okay, so it's only you.
Speaker 1:You think that in the development phase of any AI project, it will take longer than actual testing of the POC or any solution.
Speaker 2:Yes, so it's like scalability takes much more time than building the POC, because POC can be made for one or two people or maximum 10. But when you're making it scalable for millions of people, it's where you get a lot of challenges.
Speaker 1:Okay, during those AI solutions that you created, have you considered any security aspects of securing the modules or securing the AI itself?
Speaker 2:Yes, definitely. So I won't say the bad part about, but, uh, the part that azure currently messages. When you use any of the azure services, what happens is you will see in the network tab it shows that, okay, this is the api which is being called and there's no way to actually uh protect it. So when you're implementing any stuff and you don't want the other person to actually be able to see, you need to hide it or abstract it out there using some classes, and that's where the first security thing comes. Second, there's a great feature, again by Azure AI services, which is prompt shielding. So I would definitely ask all the viewers to go ahead and check out the prompt shield feature so that you can actually go ahead and shield your prompt in scenarios where you are making the prompt take important decisions. And yeah, this too, I feel, are the most important security features that should be taken care when using any LLM model or AI services out there.
Speaker 1:Okay, how are you creating these AI solutions? Are you using any SDK or are you just writing code, Angular code and just pushing it while other teams are creating in crcd?
Speaker 2:so the thing that works is, uh, I currently use a combination which is like cli, sdk and also the direct ap calls that we get. So some of the azure ai services are available like a direct api service which you can actually go ahead and deploy and then directly call them using the model deployment feature. So I use all the three or maybe I use the Azure ML studio where we can actually go ahead, use our own data set to actually train the model using maybe AutoML designer or notebooks. Use them to actually wait out there and then deploy it using the model using maybe AutoML designer or notebooks. Use them to actually wait out there and then deploy it using the model deployment feature that's out there and use it like an API.
Speaker 1:Okay, so let's dive into your community work that you've done in AI aspect, because I saw that you've been quite active. So what are some of the community things that you've been involved with?
Speaker 2:So currently, to be like a surprise or achievement, I have around hosted in one year around 55 events till now where I've been a speaker or a judge or a mentor out there, and it's mostly about AI or Azure.
Speaker 2:So in that particular event it's like the most events I see that, okay, the students are not able to grasp what is AI or understand how maybe Azure AI can work, or GitHub Copilot or GitHub CodeSpace can be used to actually boost their development, boost their whole speed of actually building the project out there or turning their ideas into reality. So my community work usually involves around all of this so that I can empower them, I can help them understand how can different things work, how they can actually go about involving ai into their daily tasks, automating stuff that's not necessary, so, like if they are writing a repetitive code, how they can use copilot to actually write the repetitive code instead of writing it again and again. How they can actually use code space to get the environment ready instead of going again and again and creating the environment when it's not required to be very honest. Or how they can build an AI service which they need, which is already available in Azure instead of actually building it from scratch using the dataset.
Speaker 1:Okay. So I was going to ask you where is your community? Is it mostly online or is it through a meetup? Can you explain a bit more for those who are interested in joining?
Speaker 2:Okay, so I have a community of my own named DayWorld, and it's online, but mostly my community work involves in different communities. It's both online and offline, so I go to different colleges, different community events out there that's happening all around, and you might mostly see me in india, uh, contributing to maybe delhi, kolkata and chennai region. So these are the major regions where you will find me contributing and collaborating with different communities, like AWS Azure Developer Community, which is a famous community in India and recently talked about by Satya also in his speech.
Speaker 1:Yeah, are you an MVP, microsoft MVP in India? I can't remember if you are or not.
Speaker 2:I'm not an MVP, but yes, would love to get a recommendation from you.
Speaker 1:Okay, yeah. So last bit is like what are things that interest you in AI?
Speaker 2:So things that interest you in AI so things that interest me, and I'm pretty sure it interests everyone is how the space is actually evolving day by day and how it truly empowers you to innovate different stuff and how it helps you go beyond the boundaries that it's already set. So let's suppose it was three years back. No one could actually imagine that we would be sharing our day-to-day activities with a, with a like with a phone that's actually giving a response more like a human, and you will feel that only someone is there. So it's all about now. You just have to imagine, you just have to think out of the box and AI can help you actually build upon it. So it's like you're getting a magic wand which can help you just do Shakalaka, and things would get there so that you can use it for different purposes.
Speaker 1:Yeah, it's the same. I find a lot of people still getting confused on AI even myself getting lots of confused and fast moving of innovation of AI with agents now. So it's a little confused, but I think the idea is try to break it down as simple as possible, like for other people to understand. So you know, as this episode is coming to an end, we love to get to know our guests, so do you have anything that you love to do in your spare time? Any hobbies, interests?
Speaker 2:So my hobbies and interests mostly goes into again like exploring and contributing contributing the community out there. So it's mostly interacting with people out there, understanding how things are going, or maybe just learning about the past events, like currently. I truly admire Steve Jobs, so whenever I get free time I try to explore how his mind worked, how he thought of innovations, how he actually spoke. Uh, what are the things that he actually did while he was thinking of a particular product? Maybe macintosh, maybe the next that he built, or like what was going on in his mindset. Or maybe other people out there, like Mark Zuckerberg or Bill Gates. What is the thing that brings out the innovation out there?
Speaker 1:Okay, thanks for joining this episode, harrison. So it's good to know the amazing work you do in AI and some of the experience that you've taught it's. Hopefully people can learn more about it. The amazing work you do in AI and some of the experience that you've taught it's. Hopefully people can learn more about it or to learn more about AI. So do you have any advice that you recommend for people that's getting started in AI or not sure where to go and things to get started?
Speaker 2:Yes, definitely, and before we end up, I would like to ask you also a question. So, first of all, coming to answer of advice, uh, the advice that I would give is, first of all, try to understand, uh, what is ai and where to use it. It's not necessary that, uh, for each and everything that's out there, you have to use AI. Some things can be done with a simple logic. Some things, which does not require a large or general understanding, can be done using ML. So it's pretty important to understand what AI is, what LLM is, what prompting is, and then go into deep into the scenario to understand the different stuff. To get started, maybe you can check out Microsoft Learn the Azure AI Fundamentals course and then go to Azure AI Engineer course. That's pretty amazing. It covers almost all of the things that's out there, from what is AI, how to use AI, and then the Azure AI service is part of it, so you can go ahead and explore that.
Speaker 1:Yeah, I also want to add that you can also check on the Microsoft AI for Beginners on GitHub, so that's a lot of tutorial for people. The Microsoft advocates created for people to learn, so that's an amazing resource, so check it out. Uh, thanks for joining this episode, harvison. It's good to get to know your stories about ai. Thank you, thank you.
