Industrial machinery and AI – Episode 6
Chris Pennington: When we consider capabilities like Gen AI and Agentic AI, where does Mendix fit in into the workflow of industrial manufacturing?
Subba Rao: As I was saying previously, I’ll look into like let’s look at some of our manufacturing scenarios. , in the industrial manufacturing like as the machine is getting used on the shop floor, there’s a lot of data that gets generated because the machine has got different motors, sensors and moving components within that. And that information is very critical for the timely maintenance and the quality of the machine.
So, what I mean by that is with Gen AI, the let’s take example of our one of our customers: Vivix in Brazil. They’re a glass manufacturer. There is a lot of furnaces, they use a lot of machinery in in their processes. Like any product, as you’re building you get some issues and that might be related to the people or process or even machine. Historically it used to take long time for Vivix to resolve the product issues. In some cases, it used to take couple of days even in the production issue. They began leveraging Mendix as an as an application layer where inspectors actually interact with that, they built using Gen AI capabilities.
Underlying is AWS bedrock as a model with all their production data so that they have a clear indication in terms of what are the issues and the resolution time. Like the time to get to the resolution has been reduced drastically by having this Gen AI based a virtual engineer. They call it virtual engineer as an application that is helping the engineers to query the information and get to the root causes and very in the typical of their hands.
It used to take few hours to days. Now it is less than an hour for them to even resolve and try to identify the production issues. Same thing on the on the field issues, because once the product goes into the field, there are some issues that might be coming. And that also is leveraged and simplified with this virtual engineer as an AI enabled tool for them that accelerates the customer satisfaction and quality of the product. And the process improvement is all the different levels that help them.
This is just one example on the manufacturing side. And in the similar tone there are others that we see taking it further with the digital twins where one customer will use the combination of simulations with the real sensors and virtual sensors. The data from the real sensors combined with the virtual sensors to simulate and give that feedback of when a certain process is not adhering as per the original intent? these are some of the examples where like that the combination of AI and Gen AI are used in customer scenarios.
Rahul Garg: Yeah, perhaps Subba you mentioned about virtual engineers and virtual simulations. The way I guess I also look at it is Gen AI today can help you build a lot of software as well, but in the context of an industrial application, you need to make sure that the software is trustworthy and can be relied upon from an engineering and manufacturing perspective as well.
When many a times you are using some software applications that’s been created through Gen AI vide coding for example, you are not 100% sure. You still need to have that assurance that what you’re building is gonna be reliable and trustworthy and I think that’s where Mendix can really help a company and outshine what companies may be thinking about doing with the simple vibe coding applications.
Subba Rao: I totally agree. Vibe coding is just one aspect of the Gen AI to generate the code. And for us in the industrial side we are looking at what the real problem is that we are solving for the customer. Generating the code means today with vibe coding you can generate the code and bring some application. But is it scalable? Is it secure enough? Is that code manageable to connect with other data that is very critical in across the landscape within organization?
What I mean by that is like in Gen AI, yes, code generation is 1 aspect that helps you for your application or even getting the context of the data but governing that and also bringing that further into your landscape—that’s very critical. And generating is not just like from the end user and the user like the process perspective. It is the generating meaningful information for the user that is not the code. It might be a graph of a certain temperature thresholds. Thresholds of a specific furnace or like quality defects in certain die when they’re using in a pressing stamping machine. These are all the different things that are very much process and industry specific that is also generated based on the data.
Rahul Garg: Yeah, yeah, I think that’s a good point. Having the ability to bring in the right process and the right context for a user and bringing that with the right data and making sure that the application and the capabilities that you’re trying to get from the application are going to be useful is where Mendix can really shine very well.
Chris Pennington: Many enterprises have invested in significant enterprise applications. Can Mendix be used to enhance them as well?
Subba Rao: Absolutely. That’s a great question. If you look at a lot of manufacturers and how they built solutions. With AI, it’s always should be looked at as how you can augment that further. AI brings more flexibility in terms of like adding on top of your existing landscape and that’s the beauty with Mendix as well. You don’t need to disrupt your existing IT landscape infrastructure and all the applications that you have built, solutions that have been built to fulfill your processes. You can add AI capabilities to it.
Just to give you an example, one customer built an application that’s collaborating from engineering to suppliers to their suppliers. And then in that scenario originally when they built it, it is there like Gen AI has been in the last couple of years. It is not that mainstream, and it’s talked about heavily. Now there’s a lot of potential that customer is seeing, and they say, OK, we already have this application that is taking the engineering information and the collaboration with supplier how we can provide some more AI capabilities into it?
On top of that same application with Mendix capabilities and also the AI advancements, they build simple assistance within the existing application that users can use to query the suppliers or the scorecard information or what are the timelines that supplier is projecting which might be spread across in different emails or different portals. It’s simplified into this existing application with the same interface so that users don’t need to go into multiple systems to even query those type of simpler things that can be solved with AI.
Rahul Garg: Yeah, you know that’s sometimes when I think about this, typically you have many enterprise applications that you would consider as a system of records, typical ERP system or a that for financial system or a PLM system for maintaining the tight controls on the functioning of the organization, which are very much necessary. But inside those applications you still need the flexibility of having the information more easily available. And I think Mendix can be a very powerful source for enhancing the value that is extracted from those applications.
Subba Rao: Yes. And you can see its value from those applications and also today the disparate, siloed nature of those applications. In at least in some customer scenarios, they’re still siloed. And with AI you can remove those silos further by having this AI enabled application that connects those different systems and get that intelligence in a more practical use case.
Chris Pennington: Subba, how do you see agents interacting with one another?
Subba Rao: That’s a good question. as I said earlier, agent is something like that’s continuously learns, monitors the data, and also takes actions. There is definitely a huge potential in terms of like removing all the manual processes that happens and having agents that can automatize those and between those steps. You can have agents talking to another agent to advance the automation. Now, let’s slice that further and look into some use case.
I was talking earlier about briefly about maintenance use case where there’s a lot of data that’s get getting captured or monitored from the sensors or mission information and based on certain events or certain data points and historical information combination you can look at users can get an alert. Typically, what happens is there is an alert then your human has to intervene and look at that alert, investigate that and based on that they have to decide whether they want to raise a maintenance ticket. A maintenance ticket will trigger a maintenance person to pick the ticket and maybe create a maintenance order and take it to the next steps. All these are manual steps.
If there’s an agent automatically creating an alert and then triggering a maintenance ticket, these are two manual processes with an agent maintenance agent that can take that action. And then there is another agent which creates a maintenance order. A machine has got manuals that are already coming from the manufacturer. And then there’s a lot of historical information in terms of how maintenance was done or how the maintenance technicians have taken the steps to repair certain things or fix certain issues; that historical information is there.
Using Gen AI, you can generate the instructions for the users. That can be like automated. An agent can generate the instructions as a first step and that can go as part of the maintenance order. These are all different individual agents that can do that task autonomously, but to connect them, yes, if it is fully automated process, you can think of connecting. But in the industrial context, I always bring the human in the loop aspect because there are certain steps where it’s critical for humans to intervene and see if it’s right time for us to approve this maintenance order. All the data points are there to trigger the maintenance order.
And in the first steps, when the work instructions are all the instructions are generated and it can even like automate that. In this case there is an agent, then the human in the loop and then agent. But in some cases, there can be agent to agent communication where that automation is simplified.
Rahul Garg: Yeah. With this whole agent to agent communication, I guess that’s the new emerging area is there some ways to simplify that whole process because you’re talking of 1 automated system connecting to another automated system? How? How will that process be more seamless or made so simpler so that you’re not running into another set of integration issues?
Subba Rao: Yeah. This is an evolving space to be frank. The agent orchestration is requirements require two things. One is actually building an agent. A simplified way to build the agent and that’s where Mendix is bringing new capabilities as well. You can build an agent in Mendix today? There are like some sample apps and examples available on how to build the agent which underlying obviously using in some cases some large language models, in some cases it’s leveraging the next generation protocols. But agent to agent communication is evolving.
And the agent orchestration is a critical piece of that one and that we have to make customer, we don’t have like at least on the Mendix perspective, customers are directly getting into that. It’s an evolving space where this agent orchestration tool will help in the agent-to-agent communication. That’s part of the like evolution that we’ll you will see in the Mendix in the coming releases.
Rahul Garg: Yeah, I think beyond just the Mendix coming releases, as you said, this is an evolving field. Many companies are trying to figure this out as well and how best to orchestrate that is going to, I guess standardized to a certain extent, with these agent communication protocols. It becomes a little bit easier.
Subba Rao: That is true and that is where we will see this MCP model context protocol as an upcoming one. It’s getting a more standardized. And that is getting more leverage. You can consider that as a universal translator for all the AI tools and that really helps in this next evolution of agent-to-agent communications.
