AIRS ML CEO Prateek Tripathi on edge technology, machine reliability, and why you can’t take a robotic arm to meetings

On factory floors with hundreds of machines, preventable mechanical breakdowns can cost companies millions. 

AIRS ML is building edge solutions that allow teams to easily monitor and maintain machine performance 24/7. 

We spoke to Prateek Tripathi, CEO at AIRS ML, about how this technology could change the face of manufacturing forever.

Hi Prateek! Can you explain AIRS ML? 

Sure! So imagine you have a car that breaks down. What is the cost associated with that breakdown? When I say cost, I do not just mean the cost of repair, but the cost of maybe not being able to make it to this meeting or of missing your flight. 

Now imagine that instead of you, it's an aviation manufacturer, and instead of your car, we have a CNC milling machine [computer numerical control machine where cutting is controlled by code], and you rely on this machine to create thousands of parts every day.

But alas, it's a bad day and something goes wrong and it breaks down all of a sudden. Well, the cost associated with such breakdowns for a single hour for one plant is nearly $1.34 million.  

We're building something we call edge processing systems that sit right next to the machines. You look at data from the machines 24/7 and run a local AI model on them. That then helps you to keep track of the health of your machines. We also have a conversational AI which is a small language model that sits on your factory floor and your personnel can talk to it and figure out what to do.

Wow! Have you already been talking to potential users? 

Initially, we got a small grant of £50,000 from Innovate UK and did a POC at the MTC [Manufacturing Technology Centre in Coventry, UK], which is one of the catapults in the UK for the manufacturing sector. We saw some good results and then we took those results and started speaking to a bunch of aviation manufacturers.

Now, we've cleared stage one of the ATI [Aerospace Technology Institute] SME grant where we are working towards a £1.5 million grant and we have a potential customer on board the grant application with us. And we have two project partners — the AMRC [Advanced Manufacturing Research Centre] and the MTC — who are writing the grant with us.

In terms of other traction, we also have letters of support from GKN Aerospace and NCMT which is a supplier of CNC machines within the UK and to the aviation aerospace sector in general. 

We also have some other ongoing conversations but that’s all under NDA for now. 

Is this idea part of your PhD studies at Imperial College London or where did it come from? 

I worked at Qualcomm before starting my PhD, specifically in the voice and music team, designing chips for earbuds and headphones. When COVID hit, I wanted to pivot to something more related to diagnostics, which led me to pursue a PhD in sensor design for things like pH detection. I also incorporate edge computing in my PhD work. However, I realized that the time to market in the biotech sector can be quite long, especially with regulatory hurdles like FDA approval, which can take up to 10 years.

Around this time, I reflected on my dad’s career. He worked in the oil and gas industry in reliability management. I remember how he would get calls in the middle of the night, sometimes needing to rush to a plant to check on machines. This got me thinking about how edge computing could be used for real-time machine monitoring and predictive maintenance to prevent those kinds of urgent, disruptive situations. That's how I started exploring this space and the potential applications of Edge AI in industries like manufacturing and aerospace. 

Nice! So reliability is becoming a family business. Where are you now in the process? Have you already developed some of the sensors or have you got prototypes that you're using?

Yes. So for the robotic arm we had our own edge system that sat on the robotic arm itself. And right now we are working on a simpler POC device.

We are also working on small language models that can do very basic stuff without the internet. It's essentially a stripped down version of chatGPT just on your laptop that you can talk to. And it has very limited knowledge. It's trained for specific tasks, but it can still converse with you and give you insights. 

So we are doing an end-to-end MVP which is easier for investors and for potential customers to interact with. Because you cannot carry a robotic arm around just to prove a point!

Haha! Do you have any particular milestone you want to hit during Techstars?

Oh, yeah. What we are really pushing towards is lots of letters of intent and getting pilots going. We have seen significant traction from the industry.

We want to have three to four pilots in the pipeline by the time we finish. So that should help show significant traction when we’re fundraising which we want to do in parallel because we also need to raise some money for the grant because grants need match funding. 

Governments want to see whether the technology will be supported post government funding and reach proper commercialization.

So, in this space, are there any technologies or approaches that have excited you?

I think edge computing will change things quite a lot, especially machine learning at the edge and tiny machine learning solutions. Mainly because the amount of data that you need to generate for monitoring an entire factory floor is unsustainable. Imagine if you want to collect data 24/7 for an entire factory floor with so many machines.

We are talking about terabytes of data. And so hopefully we at AIRS ML can convert those terabytes into kilobytes. If we can make it work, I think we would change the industry.

We’re sure you will! Best of luck.