Scott Keck-Warren
0:05
Hello developers, and welcome to the PHP Architect Community Corner, where we have conversations with members of the web development community. I'm your host, Scott Keck Warren, and today we're talking with Kumada Sharini Vasa about her talk at JS Tech. Kumada Sharini Vasa is a senior data architect at ATC Drivetrain and is the founder of Unimunk, an AI-powered education platform, and GoCure, a digital-first medical tourism solution. She designed hybrid cloud architectures spanning AWS, Azure, Kubernetes, and big data pipelines to support middle mission-critical operations with a focus on scalability, governance, and predictive maintenance. She holds a master's in data science from the University of Texas at Dallas and specializes in distributed systems, MLOps, and secure data engineering. Thank you so much for finding some time to talk to us today.
Thank you so much, Scott. Yep, it's an honor to be with you and also honored to be talking in JS Tech. That's a wonder—
Scott Keck-Warren
1:05
yep.
That's a wonderful opportunity, which we have got to have a podcast with you.
Scott Keck-Warren
1:11
Yeah. And so, and I, so I don't know you very well, like your, your name popped up because you're gonna be one of the speakers at JS Tech this year, and I'm looking forward to meeting you in person. But like your topic kind of sparked my interest. So can you tell us a little bit about your, your talk at JS Tech this year?
Sure. So mine basically is AI refactor, like how I upgraded from 20-year-old enterprise systems to modern JS architecture. So my background as a senior data architect, we work on a lot of automobile stuff where you have like 30 years, 40 years where they use PHP and like critical, you know, unique data connections, basically using InfoFlow or ERP system. But the main challenges right now is like most of the legacy systems are like need to be upgraded in Python or something, but the most function need to be in PHP and call and automation. That's what is required. So the whole thing running on a mission-critical operations, uh, I try to do a large-scale manufacturing ERP application, so I try to convert the classical ASP pages, VBScript, you know, all those ODBC lookups and printer-based systems. In this section, I will walk through, like, I would say I would break down our monolithic legacy applications and how I built it basically using the micro— JavaScript microservices ecosystems. So I'll be showing how I migrated from a VBScript-driven data collection flow into Node.js services and how the ERP queries into Python micro APIs. So if you have an old system you want to bring back to JavaScript or using new LLMs or Python APIs, that's what I'm going to talk about the whole thing and see about like what are the steps you have to follow, what are the sequences, what are the architecture, how AI helps to validate the logic when the production is still going on and to make it stable. That's my—
Scott Keck-Warren
3:17
So that's, yeah, no, and that all sounds really fascinating. Like I, I unfortunately like cut my teeth with VBScript and ODBC connections. So I'm sure that like I'll be able to get something out of that and not be completely bored by it. Um, but I do love like this concept of, cause I, I treat like AI and large language models as kind of like, um, like a junior developer, right? Like I give it like smaller tasks and then like let it take care of that. Is that, is that kind of like what we're expecting to see, um, like within this presentation?
Yes, kind of like I have done a smaller application, LLM applications, like where it validates, uh, like, you know, validates logic and production stable. So basically AI is something like, you know, a developer who's helping you out just to make sure you are trying to move the architecture when you're doing most of the work.
Scott Keck-Warren
4:08
Okay. That's really fascinating. Cause I, I, I like, I am, you know, I have like mixed feelings about large language models, but like, I find this to be really fascinating. We'll be right back after this word from our sponsor. As a programmer, you know that code reviews are critical, but are also time-consuming. CodeRabbit acts as your AI copilot, providing instant code review comments and potential impacts of every pull request. Beyond just flagging issues, CodeRabbit provides one-click fix solutions and lets you define custom code quality rules using AST grep patterns, catching subtle issues that traditional static analysis tools might miss. CodeRabbit reviews 1 million pull requests every week across 3 million repositories and is used by 100,000 open source projects. CodeRabbit is free for all open source project repos. Get started today at phpa.me/coderabbit. That's phpa.me/coderabbit. So, so, so this is something that you've actually done like in, in, in a production environment, or is this just something that like, have you done as like a testing kind of idea?
No, we have done it in a production environment. We took almost 4 months to finish off the whole architecture. The first thing is like. You don't want to touch something which is like, you know, very, very old and the legacy systems are really old. And I don't know if you have heard about InfoFlow. It's a Unidata ERP system where most of the remanufacturing industries use. So when you're touching that and you have this ODBC queries, this legacy systems and everything. So we took like the 4 months to do it into a new language on Python APIs and LLMs was basically just to validate the logic. That's all.
Scott Keck-Warren
5:46
Hmm. Oh, okay. So, so like the, the, you did more of like the, the, the, you did the coding by hand and then the large language models like verified your work?
Scott Keck-Warren
5:55
Oh, that's even more interesting. And so, so like, how, how did that kind of like, uh, manifest itself? Was it, was that part of like your code review process or, or did you like run something locally to like, to validate that the changes were correct?
Uh, basically when you, I mean, when you're using an LLM model, you generate like human-like responses, analyze, and summarize. So my problem was like, I took like small, small parts and you needed to like, you know, validate a few things with the logic because you have a lot of test cases and everything. So it was just doing like explaining the legacy systems, what was the logic and testing it out. That's how basically.
Scott Keck-Warren
6:34
So you kind of treated it like a documentation helper?
Scott Keck-Warren
6:37
Yeah. Oh, okay. That makes sense. That's really a fascinating use case for that then, because I hadn't thought about that. Yeah.
And like, I mean, like most of the architecture was converted into Node.js architecture. So LLM was just like something which helps out saying like, hey, I'm there, you know, I'm just analyzing the code and just the VBScript and that's all.
Scott Keck-Warren
6:59
Oh, okay. That totally makes sense. That's like another, I always find there's so many ways to use large language models for like coding, right? And this is just like another weird, well, not weird, but another interesting use case for how we can do that. So just as like my own curiosity, like how is it that like some of the data, I'm not a Node person, so this is probably my ignorance, but why is it some of it you did in Node.js and some of it you did in Python?
Yes. I mean, my problem with using Node.js was like I needed to do a lot of backend systems and APIs, and it should be lightweight because when you're going to the production, it needs to be like migrating like from ASP to rebuilding Node.js. Validation ERP calls was like the best form for Node.js. So the endpoints handling, validation, service logics, that's what made me into, into the Node.js architecture. So we had a lot of data collection questions. I mean, basically when you're in a production line, you have like a lot of rows just coming on and you have protocols and everything. So I needed to act as a bridge between the old ERP system and the new frontend systems. So API layer, Node.js won for me.
Scott Keck-Warren
8:14
Okay. I think that'll— it's always great to like have a combination of languages for these types of things. Some things do work better in some environments than others do. So, so when people come to your talk, is there anything that they're hoping that you'll, they'll come away with, um, as like the big picture of, of what you've done?
I would say like if they know a bit of their ERP systems or legacy systems, that would be helpful for them to understand more in depth. But I will be explaining from the basics like, oh, this is this what ERP system is. So from this basics, because so many people don't know, like ERP systems, or like, uh, what you're using in the microservices or lightweight, like why are you using like Node.js instead of PHP, or are you using PHP? I am using PHP in some part of it, but not the whole thing. So I would say they would carry like how do you design, which system to select, and to really see the, you know, good example in the legacy systems and how you migrate it from a VBScript to a Node.js basically.
Scott Keck-Warren
9:20
Okay. And I'm really curious about the VBScript to every— to something modern, right? Like, I know like VBScript still exists and people are still using it, but like, oh man, I just have nightmares about working with VBScript. So before we like, we do appreciate your time with us today and you being one of the speakers. Is there anything that you'd like to direct our listeners to before we let you go for today?
I would say like JS Tech has a lot of talks and you would go into the speakers and look at their talks, learn more, understand, maybe interact with them in the LinkedIn and like, you know, Go out, speak to them when you're in person, and also come to PHP Architect. I saw a lot of your podcasts and it was in-depth also. So I would always do that, both of it together.
Scott Keck-Warren
10:04
Yeah, I think there's some like really amazing talks this year. I like it. It's really nice that like there's going to be so many talks, but like, I'm really having a, like a struggle going, which one of these am I going to go to?
Scott Keck-Warren
10:15
There's like, cause it's like, sometimes there's like an hour where I'm just like, I want to watch all of these. And it's lucky that we're like, we're recording them. Um, but that's also a good idea to like go out and find people's LinkedIns and like connect with them that way beforehand.
Yep. That, I mean, I, I was like, I was looking into that one of the talks and I saw like Nate Finch, he had decoupling with your CMS with Next.js. I was like, I want to know more about it, you know, like that. I saw it's better to reach out to people, do a network and learn even like, you know, the basics before you come out there. Yep.
Scott Keck-Warren
10:47
Yeah, absolutely. So I do again appreciate like your time with us and I'm looking forward to meeting you in person. So thank you very much.
Yep. Excited to meet you also in person. See you next month. Yep.
Scott Keck-Warren
10:56
I have to say another heartfelt thank you to Komada for all of her time and also for yours for listening. If you're not already a subscriber, make sure you subscribe with whatever your preferred method is so you can get the episodes as soon as they're released and maybe share this post with a coworker so they might be interested in going to JS Tech as well. This podcast is available both in an audio version on our website, as well as a video version on our YouTube channel. This is Scott Keck Warren for the PHP Architect Community Corner, signing off and reminding you to keep listening, keep coding, and keep reading.