What's Up with Tech?

Inside Tupl’s Explainable, Agentic AI Transforming Global Telecom

• Evan Kirstel

Share episode

Interested in being a guest? Email us at admin@evankirstel.com

Telecom doesn’t need another AI science project; it needs outcomes. That’s the thread running through our conversation with kishore bobbarjung from Tupl, where explainable, agentic automation is delivering real wins: faster resolution of customer issues at a US tier one and double-digit energy savings across more than ten European networks—without sacrificing KPIs. We walk through how connecting customer signals to network data shrinks time-to-fix, how policy-aware power optimization lets radios sleep smart, and why trust and transparency with engineering teams are the real force multipliers for scale.

We dig into the design choices that matter: building AI around workflows engineers already use, surfacing every step so fixes are understandable and safe, and promoting repeatable investigations into closed-loop automations. That’s how costs stay predictable and knowledge compounds. We also unpack the tough parts—workforce concerns, governance, security, and the surprise bill that can come from token-heavy agents—and show how Tupl contains exposure by localizing processing and turning one-off agent work into durable scripts.

The roadmap gets bold: a “no tools” North Star where an agentic co‑pilot interrogates data, guides deep investigations, and then codifies proven remedies back into Network Advisor for ongoing, low-cost execution. With AI moving deeper into the RAN, we explore early gains in radio design, optimization, and critical-site resilience, as major vendors push intelligence closer to the edge. If you care about cutting OPEX, boosting customer experience, and replacing tool sprawl with a coherent AI layer, this conversation lays out a practical playbook.

Enjoy the episode, then subscribe, share with a colleague, and leave a quick review to help others find the show. What would you automate first in your network?

Support the show

More at https://linktr.ee/EvanKirstel

SPEAKER_00:

Hey everybody, another fascinating discussion today with a company that's making AI and automation real for global telecom operators at Tupal. Kishur, how are you?

SPEAKER_01:

I'm great, Evan. How are you? Great to see you.

SPEAKER_00:

Great to see you. You're in uh rainy wet Seattle, but uh I think that's par for the course.

SPEAKER_01:

Beautiful is part of that.

SPEAKER_00:

But uh before we dive into all things tuple and the amazing results you're driving, maybe introduce yourself and your role. And how have you described Tupol these days?

SPEAKER_01:

Well, let me start off with uh brief introduction of myself. I've uh been with Tupal uh for just about a year, coming up, I think literally next week. But um I've been in the industry, especially uh in the wireless industry here uh for about 25 years or so, aging myself. Actually, a little just about a little bit more than that. Predominantly spent my time working with uh ATT wireless, a lot of times with mostly as an outside consultant and doing a lot of work uh for uh their network design, network optimization. Um one of the things that we did back in uh 2010 or so was to uh part of uh uh my old colleagues, we kind of started off a company which started doing automation. Um at that time we didn't know it was actually AI. We used to call it automatic intelligence correlator. That was our uh our go-to slogan for what is today AI. But of course, it's it's nowhere uh near today's uh level of uh, you know, it's I guess AI on steroids at this point of time. But that's uh something that we were doing uh quite a bit and were very successful uh in the 3G era to do network automation. We're basically taking a lot of data streams, comparing information, looking at anomalies, looking at changes, and recommending uh a course of action to be taken. Right. So long uh story short, about uh a year and a half or so ago, um I was talking to uh a friend of mine who works at AT ⁇ T, and he happened to be also a friend of uh Petri, who you spoke to, I think uh a couple of weeks ago in Boston. Uh and of course, we are based in, I'm based in Seattle, I've been here for 20 plus years, and um it just happened that we were talking about it, and he started explaining to me about what Tupel was doing, and I thought, hey, this is to me, it's almost like deja vu. It's uh like wow, we had thought about all these things in the past, and we kind of just completely uh gelled together, and it was a great uh experience just talking to him about two hours uh about what Tupel does, and and here I am. I ended up saying, okay, I'm gonna work with you guys, and this is exciting that uh I'm back in the automation journey for uh network operations, network planning, you know, all kinds of stuff.

SPEAKER_00:

Yeah, and and it's been a long journey, and there are lots of different approaches to automation, but it seems like at Tupal you're taking this to an entirely new level, delivering kind of results that have been um, you know, not achieved with traditional automation approaches. Maybe describe, give an example of how Tupal is helping a global operator deliver real tangible results today.

SPEAKER_01:

Yeah, I think I can take two examples very easily, uh, which is uh again, going back a few years, Tupal has been doing AI, uh, you know, the traditional AI machine learning model for network automation, for automation of tests and stuff like that for the last few years. But the AI industry and AI itself has seems to have caught up and more become more relevant, right? Uh, but just over the last few years, we have uh helped uh tier one carriers here in the US, at least definitely one tier one, which became the number one by doing a lot of automation uh in their network, connecting customer issues to network issues, and being able to quickly resolve uh customer issues so that the level of customer satisfaction is super high, right? And I don't have to name it, but it's so it's based here in Seattle, and they are the number one. And then there's a lot of um uh proof that what we have done literally saved uh them a lot of dollars and also gained a lot in terms of uh customer uh satisfaction and the customer experience is just sky high, right? So that's that's one. Number two, uh second example, uh global carriers was is just this year um we have uh deployed our power savings advisor uh in over 10 uh countries in Europe, right? Wow. This is it's a power is a major uh expense for most carriers, and especially in Europe, because they have uh somewhat older radios at this point. They have they haven't yet caught up to 5G as much as the US has. So they're using uh partially uh you know somewhat older equipment and then this 3G, 4G, 5G, and a lot of it, irrespective of what technology is it is or what vendor it is. Uh tuples uh power saving advisor helps a lot in um configuring the network so that there is least power consumption with without uh affecting customer experience or dropping of KPIs and stuff like that. Typically, a lot of times, well, hey, how do you say power? Turn off parts of the network when nobody is using it. Great, but is it impacting customer uh you know KPIs? How is that experience, right? So, and it's to be done at scale throughout the country, the national level network is huge. So it's it's been proven, and that's why we are deploying it in more than 10 countries in uh in Europe. And it's it's also being tried out here in the US in some of our largest cities, including New York.

SPEAKER_00:

Amazing. And of course, there are a lot of different approaches to applying AI and automation in telecom, different technologies, different models. Um, but yet you you really stand out. What makes pupils uh unique approach stand out from other kinds of alternatives, let's say?

SPEAKER_01:

Well, we've started from scratch. I think that's uh as I mentioned, it's it's we started off with an AIML model, which is uh which which was predominantly to take away uh all the expertise in coding and focus more on what engineers were doing in a telecommunications network, right? They didn't so we we kind of packaged it in such a way that a specialist engineer uh who is designing the network, who would attend to network issues, would easily be able to understand what his workflow is and automate that workflow. So we didn't focus so much on uh using just cool technology, but we were focused more on exactly how that process is done manually. And if it is repeatable, then it becomes also boring up to a certain extent. And also it doesn't have to be done the same way manually every single time, right? So that's that's that's what our focus was. It is what we call explainable AI. Engineers can look at it, look at the process of what's being done, and understand what's being done so that they can say, okay, next time the same issue happens, I can automate it, right? And it uh tuples OS uh generates a, for example, a script, which, if it is a repeated offender, that it can automatically fix it, right? So that's that's been our approach, that it is cleanable. And then, of course, as we went through time over the last year and a half, and especially just over the last few months, things have are changing very rapidly in the AI industry. And and we've been able to adopt and use those newer uh things like agentic AI very quickly. And nobody else in the world, as far as I know, has been able to do it yet because a lot of them are still trying that out, and ours has already been implemented and being proven day in and day out.

SPEAKER_00:

Amazing. Well, yeah, now is not the time for science projects, but practical uh results. And of course, telecom transformation is never easy. You know, our industry tends to, some would say slow, I'd say careful and deliberate in the way they work. But what what are some of the toughest challenges or roadblocks that you're seeing right now?

SPEAKER_01:

Well, a lot of times the adoption is um, you know, it's it's got uh companies weren't ready yet, but now I think most companies are ready uh from uh from a from a focus perspective. Everybody's getting into uh making sure because it's uh it's uh it's cost efficient and it's resource efficient. So obviously there is uh you know there's an advantage to that. A lot of times the uh I wouldn't say roadblocks, but what slows things down is typically what uh it would do to people, right? You know, we've always seen this, and it's it's uh uh it's uh mentality of how is it am I gonna lose my people because everything is gonna get automated? But it's always the answer is always no. It elevates your people from doing mundane things, it automates all the stuff that they just spend time on uh doing unnecessary stuff and really makes them uh you know, it it literally elevates the kind of work that they do, right? And so that's that that's definitely one thing that's uh kind of uh slower slows things down in terms of adoption. The other one is it's new, AI is new, it's uh how much information gets out uh to an AI provider, for example, governance, security are two major concerns, right? And also cost. Cost is a major concern because a lot of it um can uh get expensive very quickly, even though it is supposed to be more about increasing cost efficiencies. If you're metering and you keep uh, you know, with agents getting more and more token usage means more and more cost, more compute, more storage. All of that is something that more telcos are making sure that they do not unnecess uh, you know, with without having a predictable model of how expensive this is gonna end up being, they are just being more careful. Governance and security and cost, those are two other things apart from human resources.

SPEAKER_00:

Yeah, and those are uh challenges that need to be overcome when scaling AI, and everyone talks about scale, but it's difficult to get the balance right. Um, but yet you you scaled with global tier one operators in the US, Asia, Europe. Um what was the secret to scaling? And um, why do you think some organizations get stuck along the way?

SPEAKER_01:

Uh I think trust has been probably one of our biggest factors. Uh we have gained our customers' trust in, again, explaining and being transparent about how things are being implemented. And we work very closely with their teams, their own specialists, their own uh people who know their process the best, right? We have quite a few so-called pre-packaged out-of-the-box type solutions, but a lot of it is customizable. A lot of it comes down to uh our customers being able to utilize our platform and creating their own ways of solving issues that are day-to-day issues for them, right? It might be one engineer in one city, let's say in Seattle, that figures out, oh, this is an issue that's happening here constantly, and this is how I fixed it. And pretty quickly it can be uh repeatable, and then that can be at scale nationally because that problem keeps repeating in many different markets, in many different areas, in many different regions. And you know, then that becomes a process, it's it's automated. And we help uh in ensuring that we, you know, we work very closely with our customers and we ensure that what they're doing is implemented correctly, and we work with them very closely. So I think the trust has been one of the biggest factors and transparency of fantastic.

SPEAKER_00:

And you know, this space is moving so quickly. Um, I I know you participate in industry forums like GSMA and the PM Forum and contribute to many of the standards there. How do you keep your team aligned when the industry moving so fast? Uh it's it must be hard to keep up and stay ahead.

SPEAKER_01:

Yeah, we we do work very closely uh with all these uh you know uh global bodies. TM Forum especially has been uh very active and uh we have been uh we are we are we are a major part of TM Forum. All our uh standards uh we follow processes that are standardized uh in TM Forum. We go back and forth on you know uh agreeing to standards, writing standards. We are very actively engaged in some of these things. Uh and we were both in uh uh uh TM Forum in Dallas, for example, uh a couple of months ago, a few months ago, I think in May of this year was uh in Copenhagen. So we participate and we are very much engaged in these global forums uh to ensure that there is a standardized process of doing it, and we follow the same processes that uh a lot of the carriers, our customers, are also following. So we ensure that there are standardized processes there.

SPEAKER_00:

Fantastic. So you mentioned Tuple's been doing AI long before it was trendy uh when we called it automation. Uh, give us a peek behind the curtain as to the team. I I I know you have a huge RD bench and you know many PhDs even. Yeah. Um and I assume the team is growing and expanding as well.

SPEAKER_01:

Yes, yeah. You know, uh the growing and expansion part, literally, like I said, it's I I've been here for just about a week shy of a year, but it has doubled in that time, right? So while most of the industry in telco kind of was slowing down in the last few years because there's been a lot of uh capital expenditure and a lot of uh uh headwinds as far as uh you know different kinds of competition, fiber, cable, all kinds of different things, right? Um, but we've we've been able to have uh sustained growth and we predict this growth over the next few years uh because of our uh history uh and what we have done in the past. And also a lot of uh us, as you mentioned, our our leadership team uh is comprised of people who have worked primarily in the telecommunications industry with carriers like T-Mobile, AT ⁇ T, uh working with uh companies like uh uh who produced software who were acquired by companies like Amdocs and Ericsson. Um so our DNA is very strong in terms of our understanding of how things work right from uh the 2G era, right? I was involved in uh the the conversion of AT ⁇ T's network nationwide from IS95 to GSM. So this was 2003 or 4. That's what bought me brought me to Seattle uh when ATT used to be headquartered here in the U, uh here in Seattle. So it's it's uh it's all that DNA and uh very focused uh um yeah approach to automating things that are mundane and whatever can be automated is being automated.

SPEAKER_00:

Yeah, it's an amazing opportunity. And I've seen a little peek into the work you're doing behind the scenes, a little bit into the next few months, even. And um you're making some big bets on agentic AI and revolutionizing workflows for field service personnel and so much more. Can you give us a peek into the months and year ahead?

SPEAKER_01:

Yeah, um, especially Agentic AI, literally, like I said, it's been a few months in development because things are changing so quickly, and we've been able to implement those changes very quickly because we were already doing practical AI and machine learning. Our uh our step ups from there to identic AI has been extremely quick and efficient, right? So we are working with two major uh carriers here in the US. One is, of course, already uh implemented and deployed, but we are following it up with us with a second major carrier here in the US, and we are discussing very closely with them on how their portfolio of tools can be you uh can be agentized. And they're the ultimate North Star for the for that carrier and us is having no tools. You have a lot of data, and you can do pretty much whatever you want to do with it by creating these agents. So one of our products, Network Advisor, is going to be uh a platform on which most of the automation happens. Most of the things that we've talked about is uh is happening on Network Advisor. But what we're doing on top of that is kind of like a co-pilot, would be the agent. When you want to do more investigative work, you can go to the agent and you can start uh asking uh queries and figuring out in-depth investigative stuff using the agent, right? And once that's done, if you're uh seeing a repeatable process, that can once again be automated, but then it goes back to an automatic automated process. So there's no continuous use of tokens and it does not become an expense. Because that that is a huge uh thing that carriers and everybody else is looking at because everything comes at a cost, right? So, how can it be localized? Because the information is already there. Once it's generated, you don't have to go through the same process again because that process is repeatable as a closed loop action. So that's that's something that we have been doing, and creating uh smaller agents to do tasks that typically take uh takes an engineer a few hours to do can be automated and can be done by an agent very quickly.

SPEAKER_00:

Amazing. Well, that's quite a revolution happening under the hood, as it were. Yeah, as we head into next year, what's the big trend or a big shift that you think everyone's underestimating right now? Any uh any thoughts about 2026?

SPEAKER_01:

Yeah, this there's quite a bit of uh stuff going on. You know, we have been focused a lot on uh network operations and customer support. Uh as you've seen, uh big uh big players like Nvidia, Nokia, Ericsson, they're all entering into the fray of making AI uh in RAN, uh, that's radio access network, a big thing. It's it's uh we are kind of already getting a head start on that by utilizing all these data sources and using AI for network design, network optimization, where networks are uh potential, where they have a potential to uh fail, especially in critical areas. Critical facilities is something that we are looking at because uh you know telecommunications infrastructure is is kind of a it's a national security issue at a lot of uh times, right? So we are ensuring that critical facilities, critical infrastructure, we are using AI to make sure that they're functional all the time. And if there are issues, they come back up very quickly. So that's that's a major area of uh focus for us.

SPEAKER_00:

Well, staying ahead of the game, to be sure. Well, congratulations on all the success, much more to come. Appreciate all the insight. I I learned a lot as always. Great, thank you very much. And and thank you for joining. Thanks everyone for listening and watching and also sharing the episode, checking out our companion TV show, techimpact.tv on Bloomberg and Fox Business. Thanks, Short. Thanks, everyone.

SPEAKER_01:

Thank you very much, appreciate it.