Context, Tools, and Trust:Missing Ingredients in Many AI Strategies, with Luminovo's Timon Ruban

The AI revolution in electronics manufacturing is gaining momentum, but navigating between hype and practical applications remains challenging. In this candid conversation from APEX 2025, Luminovo Co-Founder Timon Ruban shares a refreshingly honest perspective on AI's evolution in the industry.

Timon recounts Luminovo's journey from AI consultancy to disillusionment with the technology, only to rediscover its potential with recent advances in large language models. This mirrors what many companies experience – initial excitement, followed by frustration, and now a measured optimism about AI's practical applications. What makes today's AI landscape different? "It's not that hard to do," explains Timon, highlighting how implementation barriers have dramatically fallen compared to previous technological waves.

For electronics manufacturers, the path forward involves targeted, ROI-focused applications rather than attempting to transform everything at once. The supply chain, with its abundance of unstructured data in PDFs and emails, presents immediate opportunities. Timon details how AI can extract critical information from these documents and eventually support more complex decision-making through "agentic workflows" – collaborative processes where AI makes proposals that humans can review, refine, or approve.

The key to successful AI implementation lies in balancing two critical elements: context and tools. AI needs both comprehensive data about your operations and specialized tools to verify or execute actions. Companies with established industry expertise and partnerships are uniquely positioned to guide manufacturing teams through this transformation. Rather than replacing human expertise, the future promises a synergistic relationship where AI handles routine tasks while enhancing human decision-making capabilities.

Ready to move beyond AI curiosity to strategic implementation? Discover how leading manufacturers are already capturing value from targeted AI applications in their operations.

EMS@C-Level Live at APEX is sponsored by global inspection leaders Koh Young (https://www.kohyoung.com) and Creative Electron (https://creativeelectron.com)

EMS@C-Level is hosted by global inspection leaders Koh Young (https://www.kohyoung.com) and Global Electronics Association (https://www.electronics.org)

You can see video versions of all of the EMS@C-Level pods on our YouTube playlist.

Transcript

Philip Stoten 0:09

Hello, I'm Philip Stoughton. I'm at Apex 2025 and I'm on the Creative Electron booth and I'm joined by Timon from Luminovo. Great to see you. How are you, my friend? Good to see you, phil. Thank you very much for having me. Absolute pleasure. I poked my head in the back of the EMS Leadership Summit a little bit on Monday. You had a really nice audience, a real nice mix of EMS companies, even some guys from Europe, which was nice to see. You were spending your time on the panel there talking about AI. Everybody is, I would say, ai curious. They're not quite sure what to do with it.

AI Journey: From Consultancy to Disillusionment

Timon Ruban, Co-Founder of Luminovo 0:40

Yeah, what was that discussion like? I mean, I really enjoyed it. It was, for me personally, back to the roots moment. I shared a little bit while I was on the panel that at Luminova we started six, seven years ago as an AI consultancy. Yeah, then got really sick and tired of AI, and so we're in an interesting moment now where you have many people that get tired of the crazy buzz and hype around AI and can be really unnerving to the degree of boring. Yeah, and I've gone through that phase myself, yeah, in the last four years.

Philip Stoten 1:15

Is that what Garner calls the valley of disappointment?

Timon Ruban, Co-Founder of Luminovo 1:17

Yeah, and those last four years for me personally, we didn't do so much with AI Some features here and there. It's only since the last three, four months, really, for us personally at Luminovo that we've been really digging our heels in again. I've been sharing that story and I felt there was some resonance within the audience that many people were shaking their heads when I was telling my story of getting annoyed, bored, frustrated and it felt weird for me being the AI hype boy on that panel because now I'm actually really, really excited about all the opportunities that large language models offer Tried to share a little bit of that panel and I think there was good resonance there was. Actually, before we started the panel, I spoke with Mark Wolf who moderated it and he said he was thinking that we start with a very general introduction to what is AI.

EMS Companies Exploring AI Implementation

Timon Ruban, Co-Founder of Luminovo 2:05

What is that buzzword? People probably don't know so much about it, but then we asked around the room to see like, okay, who uses ChatGPT? Everybody who uses it on a daily basis, and it's like still almost every hand up there and it showed you that people are really engaging. And how can you not?

Philip Stoten 2:22

at this point.

Timon Ruban, Co-Founder of Luminovo 2:22

Absolutely. But also in electronics, people are, I think, on a personal level, engaging with technology a lot, and where a lot of hands went down was when it came to okay, have you used it to really automate a process within your business today? And I think that's where we're in the very early days, that's where a software company in this industry like us sees that as our role, is very excited about it, yeah, how we can leverage those possibilities that people already feel, yeah, and their personal lives, yeah, and bring that to a business level as well well, excited, frustrated, a little bit angry, yeah, and now you're back in a positive frame of mind about it.

Philip Stoten 3:03

what I found really interesting this week I've spoken to a lot of CEOs and leadership at EMS companies and a lot of them have more people here than they normally have, and some of those people are here just tasked with go and figure out what we should be doing with AI. Go and figure out which partners truly understand it and have it and have something we can operate with. And try and help us find out who our guides need to be through this AI journey. Because when I look at Industry 4.0, which has been around for like 10 years, there's a real frustration with that. It feels like it hasn't delivered what it should have delivered. All we've got out of it is connectivity and a lot of the data Great news for AI, because that's kind of a foundational point. But I feel they want to have a better experience with the way they introduce AI to the business.

Why This AI Revolution Is Different

Timon Ruban, Co-Founder of Luminovo 3:55

How do you think you deal with that? I mean, the simple answer is I think they will have a better experience. That's why I'm personally so excited about it right now. I think people get used to things really really fast. Right the first time before you ever use something like ChatGPT, you think that's an alien technology.

Timon Ruban, Co-Founder of Luminovo 4:13

You use it two times and you're used to it and a little bored and annoyed that it didn't get that one thing right. So it'll be a process of some months, maybe even years, but I think, if you compare it to other technologies that have been hyped over the last two decades or so, including AI itself, five years ago, six years ago, predictive maintenance, when it wasn't at the same level that it is today.

Timon Ruban, Co-Founder of Luminovo 4:40

Those were things that happened over many almost you could say decades. With an S, I think, ai. This time, personally, I think it'll be different, it'll take a few years we're still talking about months and years, but we're not talking about a decade. That people will feel tremendous change. The reason is that it's not that hard to do. That's the big difference. In the past, you needed really valuable problems because it was really hard and you needed a lot of data to be able to do anything very expensive.

Timon Ruban, Co-Founder of Luminovo 5:10

And that's very different now. That's why there is this hype. Parts of it are still overhyped. Sometimes it can be confusing, but now it's. I mean, I don't say easy in the sense of it's super easy, but it is much, much easier. It's a little bit like writing software. Anyone can write software. You can do a good job at it and a bad job at it, and I think that's the level that applying large language models is at right now. It's actually pretty easy. Everyone can get started very quickly. You can do it in a good way and in a bad way, but it is easily doable to get something out of it and I think we will see a lot of adoption in the whole industry over the next two, three years.

Extracting Value from Unstructured Data

Philip Stoten 5:50

One of my senses is that when they look at AI and it's the same way they looked at Industry 4.0, they want some tangible, controllable, ring-fenced projects that have clear KPIs and have a return on investment, rather than trying to boil the ocean and do everything in the factory at once. And some of them look to specific pain points. They have Often those a lot of pain points that they've had recently have been in the supply chain. Some of the pain points in the supply chain relate to tariffs and current geopolitical issues. Supply chain is kind of an area that you've kind of made your home. Is that an area you think AI can really have an impact?

Timon Ruban, Co-Founder of Luminovo 6:31

A hundred percent. I mean, we personally at Luminova right now we're starting with what I would call the low-hanging fruits, and the low-hanging fruits is all about going from unstructured to structured data. Yeah, so I've been building in the last couple of weeks, kind of in my free time, in the sense of I usually have to manage the company but I get to do some hands-on coding myself.

Timon Ruban, Co-Founder of Luminovo 6:51

Yeah, I usually have to manage the company but I get to do some hands-on coding myself. I'm working on PDF extraction use cases, which PDFs that's the heartbeat and email and Excel of the supply chain, where you have offers, pcb specifications, data sheets, all that types of information that today you have.

Timon Ruban, Co-Founder of Luminovo 7:04

People look at it and read it and get the information out, and that's now much easier to have AI, look at those PDFs, extract the information that you have, and so it's a kind of, I would say, low-hanging fruit, because it's very obvious it works much better than it did a year ago, two years ago. And where it gets interesting and what I'm very excited about but I can't give you the full answer yet is when it comes to more agentic workflows.

Philip Stoten 7:30

Yeah, I was going to talk to you about agentic. I think that's really interesting. I'll talk to you about it. Yeah, let's talk about it.

Timon Ruban, Co-Founder of Luminovo 7:36

You don't have to talk to me about it.

Agentic AI: Human-Machine Collaboration

Philip Stoten 7:38

Yeah, I'm going to dig in there. I'm sorry about that, but I've been looking at all these AIs and, as I say, I think for an EMS company right now you want those controlled, ring-fenced. You do a few of those projects and then what's clear to me is you've done that in your supply chain, you've maybe bought and synthesized all your, say, your inspection data together, families on your pick-and-place machine, perhaps of components, all that, and create a agentic model that is actually interrogating all those AIs. That becomes almost the tool of management to help in that situation. Is that the scenario you're thinking of? Or you don't have to tell me, because I know this may be some commercially sensitive stuff there.

Timon Ruban, Co-Founder of Luminovo 8:31

Yes and no I would say. When it comes to agentic things, it's still a little bit up in the air how far we can go, how fast, different from the unstructured to structured, because it kind of works. Today, is it 90%, go to 95? There's a little bit of tuning and that's, if you do it, bad or not so bad with agentic.

Philip Stoten 8:55

People are kind of figuring it out as they try different things. Well, I think in a lot of cases in manufacturing, 90% or 95% are still not good enough, because that 5% is a mess and very expensive Exactly.

Timon Ruban, Co-Founder of Luminovo 9:05

I can give you a little bit of my mental framework of how I think about how good agentic workflows work and then what exactly which workflows they will be applied to. I think in the end, it's any type of human decision making when I think it'll be a long time before the AI agents completely replace. I think there'll be a long time of collaboration between agentic kind of agent. Yeah, ai is making proposals. Yeah, that humans will then double check and be to get the 95 to 100 percent right and be like is that a good idea? Why did you make that proposal? And and get that? And be like yep, yep, no, no, no, yep, yep, yep. And this, since I'm, I'm it really is that powerful that it can be any decision making process. That's why I think it's unclear at this point which are the ones that are its most suitable to We'll as a supply chain management software startup.

Timon Ruban, Co-Founder of Luminovo 10:02

We'll obviously start with the type of supply chain decisions that you have to make, because it's just a natural next step to us Things like proposing part alternatives after you read a data sheet and interpreted the life cycle and the price and got all the data together in one place to make those smart decisions that humans today make about which part alternatives to approve, which vendors to choose, who to negotiate with, what action to take next. So there's a lot of these decisions that today, within our software is basically people clicking around. That's what a SaaS software is today, including Luminobo. Today, luminobo is this platform where we try to get a lot of data in one place and then it's the human clicking, doing a lot of clicking, and we already found ourselves in a good and nice UI and you have to click a little bit less than with our competitors.

The Context and Tools Framework

Timon Ruban, Co-Founder of Luminovo 10:49

But at the very end of the day, there's still a lot of clicking involved and I think that might change drastically Throughout the industry, hopefully at Luminovo first. But those are the topics I'm looking at right now the clicks the human today has to make. Maybe in the future it'll be more of a collaboration of the AI making direct proposals and you're more going through it, and there'll be some clicking Of being like yes, no, no, no, yes, this is good. Yeah, have you looked at that A little bit more like a conversation?

Philip Stoten 11:18

Yeah, no, absolutely Than just a one-way street. I absolutely agree with you there. And when I think of the evolution of the process, first stage is that it's maybe what you would call the co-pilot stage, where it's saying hey, you should do this, and you're clicking yes or no. And if you find in a specific discipline that it's right 100% of the time for a reasonable amount of time, you'll say, actually, within the parameters of that small group of decisions, go ahead and do it. And I think that evolution is what we need. But I actually think for the EMS guys, they really need someone to guide them along that journey.

Timon Ruban, Co-Founder of Luminovo 12:03

And I'll give you that mental model I was speaking about now, like how what you need to make a genetic workflow work really well and we obviously need Dell M, but that's outside of our control, so we take whatever. The best thing right now is that OpenAI or Google and that's part of the exciting stuff is every three to six months these models are improving. But the other part to really make it work for your decision-making workflow is basically context and tools. So it's all about to have LLMs make good predictions. Context do you give it. So what data do you put in there? Because if you just ask ChatGPT about the availability or the life cycle of some part, it doesn't know.

Philip Stoten 12:47

And it will confidently guess.

Timon Ruban, Co-Founder of Luminovo 12:49

It will confidently guess something, but you can give that as good context. So it's still, and in that sense especially, the supply chain space hasn't changed at all. So that's the value of Luminovo today and it that's the value of Luminovo today and it will be the value of Luminovo in the future, when it's agents doing more of the decisions than humans.

Timon Ruban, Co-Founder of Luminovo 13:07

The value is getting all of the context of the supply chain in one space so that the AI or the human, or best both of them together, can make those decisions. So that's the context part and that's one part where I feel very comfortable about the position that we're in, because that's going to be crucial in any of the many futures that I can predict of where AI will go. And then the second part is tools, because if you think about a decision-making process like cleaning a bomb, which is a little bit of a mix between decision-making process and this going unstructured to structured data, but you'll have some text string in that bomb that to you looks like an MPN, because you're human and you know what MPNs look like. So you're going to take that text string and you put it into the DGKey or Octopart search bar and you search if it exists. And that's how people today manually clean bombs.

Timon Ruban, Co-Founder of Luminovo 14:01

With tools such as ours, some of that's already done automatically for you. But if you want to now go all the way and have LLMs do that, they don't know. They can just as well as the human guess whether that text string looks like an MPN, but do they know whether that MPN exists or not? For that they need a tool, they need to have a part search, just like the human has a part search where you can double-check in the database this MPN, this manufacturer, do they exist? Make that call, use that tool. And that's the second part that you need to make agentic AI workflows work really well context and good tools.

Timon Ruban, Co-Founder of Luminovo 14:38

And the more powerful your tools are, the less context you need, so to say.

Philip Stoten 14:44

There's two ways to solve it.

Timon Ruban, Co-Founder of Luminovo 14:44

It's a balance, isn't it?

Timon Ruban, Co-Founder of Luminovo 14:45

If you think about the MPN thing, the brute force way to solve the MPN problem would be put all MPNs that exist into the context, but that's not possible because there's millions, billions of MPNs, and that's why you need a tool in this case.

Building for an AI-Enhanced Future

Timon Ruban, Co-Founder of Luminovo 14:58

And so there's a trade-off between powerful tools and powerful context that you can cross-reference, and that really is what I think. Agentic is a little bit of a buzzword already, but that, when I think about agentic workflows versus just like a single LLM application, the difference is this interplay between context and tools, and in agentic workflows you'll iterate. You'll have an agent that has access to tools, like he can. If he wants to make that decision, he can search for an MPN, he can look up the lifecycle, he can look up the availability and the offers and take that and put it in the context of the next iteration of this decision-making process. And it's that, this multi-step decision making process between powerful context and powerful tools. That's what, in my head, is like the definition of an agentic workflow, and that's what we're working on at Lenovo, and getting the right context, getting the right tools, and we have a lot of that. That's what makes me very optimistic about the future for us, the future for us, the future for the industry. You could get easily scared, right.

Philip Stoten 16:06

Just be like oh, it's going to replace everything we've built. And now? All the clicking software that we have now is it going to be all for nothing Now?

Timon Ruban, Co-Founder of Luminovo 16:12

that we do everything with AI and there's going to be some other startup that does everything with AI, but the calming answer for Luminovo is no. To make AI engines work, you need good context and you need powerful tools, and that's actually what we have. We have good context we have a lot of the data and we have powerful tools and we're still working on making them more powerful, better context and then putting them together.

Philip Stoten 16:36

But also you have great partnerships. You have great partnerships and those partners are going to tell you what works and what doesn't work, what you need and what you don't need, and I think that's hugely valuable. You've established very quickly a very strong position in the industry, particularly in Europe, with some really major players, and I think that gives you such a strong lead in this journey that these companies need your help to be taking on.

Timon Ruban, Co-Founder of Luminovo 17:03

I think that's exciting.

Philip Stoten 17:05

Timon. Thanks very much for stopping by. Absolute pleasure. We'll talk again soon. Thank you, phil.