We sat down with Nikunj Mehta, founder and CEO at Falkonry, discussing their primary areas of focus, benefits of predictive maintenance, challenges of data integration and their founding story. Below you can find our podcast edited for clarity and brevity. Transcript is more comprehensive as we needed to leave out some sections due to sound quality issues.
<Beginning of section only in transcript>
Nikunj: I went to your website and I saw a lot of use cases. I liked it. We like use case marketing. So, I’m glad that you’re doing it.
Cem: Because whenever I talk to enterprises, they ask, okay what am I going to do with this AI thing. There needs to be a complete answer and very few people have that complete answer. That’s essentially why we launched appliedai.com and why we are now going into a bit deeper into detail.
Nikunj: okay yeah that’s terrific I mean on our website as well we put out a few use cases and our objective is for users to understand that this is real technology. It requires only a few choices and that it solves a certain type of problem. I see more AI should be sold through cases so people can find out what technologies are appropriate for their uses.
<End of section only in transcript>
Cem: Since we are both interested in use cases, let’s start talking about them. I was thinking about predictive maintenance but are there any other use cases that you feel even more strongly about?
Nikunj: Yes. certainly, we’re talking about process optimization as an area in which predictive analytics is highly valuable. We are finding that many of our customers are operating complex machinery and process to achieve high quality. For them a challenge is that the physics and chemistry they’re applying is complex and has very low tolerance margin. For example, in semiconductor, the critical dimensions are basically a few nanometers. So, when you are doing such high precision manufacturing, you must achieve high quality and your equipment is really stressed out to get that high quality. As a result, if you do not calibrate and maintain your equipment well, then you are not going to get great results. It’s not that the equipment has broken down but that you need to maintain high degree of calibration to get high quality.
Cem: So you help with finding the right calibration, I mean adjusting the machinery to the right calibration, right?
Nikunj: True. I’ll give you an example. We have a customer that is doing semiconductor fabrication and for them, it is a competitive need to achieve high throughput and quality, higher than where they are at now. One of the challenges is that as the dimensions are getting smaller, they must be even more precise in terms of when they perform maintenance. We’ll call it maintenance but it is a type of intervention. To do that intervention at the right time can have a major impact on the production throughput and quality.
Currently, they are performing this calibration at some interval, let’s say a few months. They look for a certain recurrence of the need for resetting their equipment. So, a lot of alarms get generated. The operators typically do not look for how to interpret those and so they are used to pressing the reset button a few times. After they press the reset button a few times, they’ll call the maintenance team and say I’ve already done this what should I do next? I’m still getting a whole bunch of alarms. Therefore, the maintenance engineer then comes in and looks at the alarms and says, can I let this go on for a little longer. Well, probably reset the machine couple of times themselves. After all of that is over then they will come in and do a complete maintenance which will establish calibration. Now, during that process a lot of time that equipment was shut down and was not good. That is lost. Also during that time, a fair amount of bad production that is going to be discovered later.
Cem: Just to be the devil’s advocate a bit, because we are talking about essentially very expensive and specialized equipment and isn’t this a different machine learning problem in every industrial setting? So, my challenge will be, if it is so customized, why wouldn’t the companies operating the equipmentbuild their own very custom solution? Why is a company like Falconry building a solution for multiple companies? Where does the value come from?
Nikunj: Every machine is different; its behavior is different. However the means by which you find patterns of interest is the same. Falconry has developed an engine that can automatically differentiate patterns that are arising without the need for any customization to be performed. No programming, no data science knowledge no consulting is needed to establish the baseline of problems. Knowledgeable user of the domain can label the patterns, so Falconry does. Much the same way that you tag some somebody’s picture on social media. The behavior that you are interested through the patterns that are arising in them and therefore you end up with a purpose-built pattern recognition technology without having hired any consultants or engineers or data scientists. So, the reason why does it make sense because otherwise it would cost 100 times that much.
Cem: That’s amazing. Another question is how do you get the data from the machines? What I have been hearing about is that some of these machines are using legacy systems and some of their data they are not even recording and just deleting as they are progressing.
Nikunj: These are all real problems. This is the other interesting thing that we are observing which is that customers want to prove that analytical technology for their problems exists. Once they prove that those technologies exist then it’s a very short-term infrastructure effort to hook up their data sources. It makes sense because they know the value. Now they can make me an investment based on the return. Otherwise, I can put out a tender or an RFP and say I have this type of data and I want general purpose middleware to connect it to any kind of application. That becomes a relatively open-ended process. Therefore, we end up finding that customers are taking a long time to decide what is infrastructure they want to spend on. The problem really is, those customers would first select analytics technology and once we have the analytic technology, they can prove the ROI and the infrastructure comes very rapidly.
<Beginning of section only in transcript>
Falconry also offers integration libraries. We help our customers cut down at least half of the integration. We work with system integration partners as well as vendors. They can create a standard integration connector that could be used by many customers. You don’t have to do this one customer at a time.
Cem: What type of manufacturing clients are you seeing the most traction with?
Nikunj: We are getting into lot of automotive manufacturing and those that use industrial robots extensively.
Cem: Because they are operating relatively expensive hardware?
<End of section only in transcript>
Nikunj: It is expensive equipment and they have tried to optimize their supply chain for the manufacturing process. Therefore, any interruptions in those production lines are very disruptive to their supply chain. Therefore, they can get even more value out of their supply chain if they had more predictability of those critical equipment. Also in many cases there is a lot of waste because of poor quality being produced from a robotic process. That also affects their supply chain because they end up producing a fair amount of waste
Cem: If I were an equipment manufacturer one concern would be that I am becoming a bit a dumb manufacturer because I’m producing this very expensive hardware but I am cut off the analytics essentially the cool software part of it.
Nikunj: Falconry is just a component into their Sales process. Then, in fact, it is better for them, because, to create this type of technology themselves can take them many years and cost quite a bit of money as well. They will be competing for the same type of people that are going to focus on learning natural language processing or doing computer vision. And talent may be interested in going to work for these robotics companies. The competition for talent is very real. Economics around this make the most sense to engage a partner like us who are able to offer them eighty percent of the solution where they can then do the contextualization without having to hire specialists.
Cem: Thank you very much for your time indeed. It was quite a learning experience for me.