Series introduction

‍

Kristen Hilley: We're asking the hard questions about AI and we're here to listen to the answers. I'm Kristen Hilley at Nymbl, and this is AI is Personal. An interview series brought to you from our Women of Nymbl AI team.
‍

Our guests are women in technology at companies across the industry, sharing what the work actually looks like from where they sit. No hype. No theory. Just real conversations from people in the trenches. And yes, AI does help us transcribe and clean up our transcripts, and help structure and draft some of our copy. However, a human always reviews all of it before it reaches you, and every voice you hear is a real, authentic one.

‍

Meet Meghna Sinha

‍

Barkha Sinha: Hi, Meghna. How are you?
‍

Meghna Sinha: Hey, I'm good. How are you?
‍

Barkha Sinha: Good, too. I hope I'm not taking some good chunk of your day today.
‍

Meghna Sinha: No, no, that's okay.
‍

Barkha Sinha: Okay. So let's start by telling the people a little bit about you.
‍

Meghna Sinha: Hello. I am Meghna, and I am an SDET at Netradyne. I have been an SDET for the past eight years now, and currently I am responsible for testing AI and models in my company.
‍

Barkha Sinha: You said models. I just want to know what exactly you work on, in very layman terms — I don't understand what you're talking about, but I should, after what you tell me.
‍

Meghna Sinha: Okay. So I work on systems that process videos from AI-powered dash cameras installed in the vehicles. So these vehicles are mostly trucks right now, and the camera detects events like speeding, distracted driving, or signs on the road. And those go through multiple backend services before they become something a customer can view. My work is to make sure that entire pipeline works correctly. That means I validate the videos. I validate how they move between multiple services, and if the material is correct, if the AI decisions are made correctly or not. And there are multiple checks here. So a lot of my work is about making complex systems behave reliably at scale.
‍

A day inside the pipeline
‍

Barkha Sinha: Okay, that sounds pretty interesting. So if we go into what a day in your life looks like.

‍

Meghna Sinha: On a typical day, my day is a mix of investigation, automation, testing, and collaboration. So I start my day by checking if all of our services in production are working as we expect them to work. If there are any problems, I go through the logs, I check the queues, I compare if the output — the videos, or whatever output we are looking for — is fine or not. Then, if there is a deployment planned, or if there are any PRs to be tested, I test them. And I have to test PRs from all of the teams. There are four to five teams, and all of them are generating code, which we are testing here. And a significant part of my day is also discussing with the multiple developers and different project managers about what we are doing, how we're going ahead, what our deployment is looking like. So the goal is not just to find bugs. It's to understand how the entire system is functioning. And this is what I do every day.

‍

What she can't stop thinking about

‍

Barkha Sinha: All of us — everybody nowadays talks about AI and what it's impacting. So you're actually looking at how it's not working at times, right?

‍

Meghna Sinha: Yes.

‍

Barkha Sinha: So is there something like maybe a problem that you might have faced and just can't stop thinking about, even when you're not at work? Maybe something generic. Is there anything like that?

‍

Meghna Sinha: So, I am a tester. For me, the biggest thing that I literally talk about or think about every day is how to make testing less dependent on manual effort. So, as I told you, a lot of testing is on videos — how it is annotated, and how the blurring and the privacy things are achieved. So this testing right now is heavily manual, which we are trying to resolve. We have automations in place, but we are still working on it. And this is something that never goes out of our mind.

‍

Barkha Sinha: So basically, even though we have so many ways to automate things, there are still things left out of it.

‍

Meghna Sinha: Yes, and there always will be. We can never achieve 100% automation. And that's the challenge, I guess. And that's also the hard truth, in my opinion.

‍

Finding the bug is the easy part

‍

Barkha Sinha: Yeah. Okay, so was there anything that you've maybe figured out the hard way and just want everybody to know — and nobody has been warned about?

‍

Meghna Sinha: Okay, so every time, whenever I find a bug, the only thing I think about is: this was easy, right? To find a bug is very easy. The only thing that's difficult is to find where it is happening from, and especially when the system is so distributed. We have a camera from where the videos are coming. We have queues where the videos are stored. We have instances where all of this is getting processed. Then we have post-processing on cloud. So there are so many distributed systems that, if I found a bug, we don't know where exactly it is happening. And that is something I had never thought I would face before. I had not worked in such a big system. And now, when I look at all of this — there is a responsibility from QA, from the SDET also, to give a finding of what the bug is. We cannot just say that this is not working. So this has been a hard truth about our job: debugging takes major time after finding the bug.

‍

How she found this work

‍

Barkha Sinha: That's definitely true. You have been doing this for eight years, you say. Was there a specific moment when you realized — when it clicked — that this is what you're supposed to be doing? Something that pushed you towards being where you are today?

‍

Meghna Sinha: There wasn't one specific moment. These things happen gradually, as I started debugging larger systems instead of isolated pieces of code. I realized I enjoy taking something that looks random — like a failed deployment or inconsistent AI output — and breaking it down until the root cause becomes obvious. So it's very satisfying to debug these things and then find out the output, and then to think about how to automate these things, so that later on this is not something that we miss while manually testing. These things gradually started growing on me. And the more I think about it, the more I realize that it's pretty fun.

‍

What the AI conversation gets wrong

‍

Barkha Sinha: Definitely. But I have one question here. You work with AI right at the front. That's the entire thing that's been going on around you. So with the firsthand experience you have, do you think there's something that the mainstream conversation about AI keeps getting wrong?

‍

Meghna Sinha: So people often think that AI is just about building models. But in reality, most of the engineering effort I have seen happens after the model is built. So you need pipelines that move the data reliably. You need the services that scale. You need monitoring, testing, versioning, privacy protections, deployment automations, and multiple ways to measure whether the model is actually behaving correctly. A model can be very highly accurate, but it cannot be useful if the surrounding system around the model isn't built reliably. So people forget that. It's just like — if you're using ChatGPT, you're not just using a model. There is so much going on around it, which people don't even look at. So I think that is something that people keep forgetting. And that is the reason there are so many companies who are making GPUs and all right now. Nowadays, it's so profitable. It's such a big deal. And it's not very easily visible to our eyes, so people just forget to talk about it.

‍

Hesitation, and using AI in her own work

‍

Barkha Sinha: I kind of agree. That's one of the reasons why people just tend to think of AI as just a chatbot.

‍

Meghna Sinha: Yeah, exactly.

‍

Barkha Sinha: I mean, you give in an answer, it gives you out another set of paragraphs. So there are a set of people who are a lot hesitant about using AI, mostly because of maybe the environmental impacts. There could be people who are saying it's brainwashing people, or that it's just making them use their brains less. It's something a lot of people talk about. So do you have something you would like to comment about that?

‍

Meghna Sinha: Yeah. I have used AI a lot here. I mean, not only testing it, but also using it to improve my automations. And I have seen how powerful it is. And it is not easy to think of solutions using AI. People would think that — oh, you have a model, right? Do whatever you want. You can create so much. But what are you going to create? The "what" is the main thing. And people don't realize that that takes effort, that takes an idea — which uses your brain, right? So software engineering itself can improve so much if you use AI — not only the development part. Obviously, we have seen Copilot and everything working so well for developers, right? Even the testing, even the manual parts — small parts which take up a day or two of testers can be reduced to an hour, like updating your test data. So we have huge test data, and updating it every month takes effort for anyone. And that can get updated in one hour by AI. Is that not good? I don't see any negativity there. I only see that my two days a month are free now to work on something else. So that doesn't mean that I'm not using my brain, or that someone is not working hard enough. You are just working on a different thing, and people forget that.

‍

On environmental impact

‍

Barkha Sinha: What do you think about the environmental impact of it?

‍

Meghna Sinha: I agree there is environmental impact, but we always overcome these things, right? And AI is not the only one with environmental impact. There are multiple things impacting the environment. Let's take chlorofluorocarbons, for example. When we were kids, we had learned that this is very harmful to the environment. But now, if you see, we don't use it anymore. So we figure out ways how to save our environment. So eventually — maybe not in the next year, but eventually — we'll figure out how this can not be as bad for the environment as people figure it out to be.

‍

Barkha Sinha: True. When the ACs were becoming mainstream, chlorofluorocarbons were definitely one of the biggest challenges that humans had to face at that time. But we overcame it, so we'll do the same for this.

‍

Meghna Sinha: That is true.

‍

Barkha Sinha: I mean, necessity does create ways, right?

‍

Meghna Sinha: Yes.

‍

A letter to her younger self

‍

Barkha Sinha: If you could talk to yourself — let's say from when you started working, maybe 10 years ago — do you feel like there's something you should tell yourself? Like a letter to your younger self? What advice would you give to yourself?

Meghna Sinha: My only advice to myself would be: do not try to become good at something right when you start it. I think this is something that everyone goes through. Once you start doing something, you want to be instantly good at it. And these things take time. It's said that you need to do something for 10,000 hours to be good at it. But I think that is where a lot of us fail, and I have also failed. That has stopped me from achieving a few things at the very beginning of my career. And I'm not only talking about work — this has been about most things. Let's take something basic, like cooking. So I would want that I'm cooking something new, and I want it to be perfect at the first try itself — which doesn't happen. And I think, if I met my younger self, that would be the first advice I'd give: take time, do something for at least 10 to 15 times, and only then expect output. Otherwise, you're just learning.

Closing

Barkha Sinha: That's actually good advice. Okay then — thank you, Meghna. It was really nice talking to you, and thank you for your time.

Meghna Sinha: Okay. Thank you.