Arjun and Kanal Jain, Building Tandem, an AI-based Knowledge Layer for Mechanical Engineers

We talk with Arjun and Kanal Jain, co-founders of Tandem about building an AI knowledge layer that captures design decisions, links requirements to CAD, and helps engineers spend more time designing. We compare text-to-CAD promises to enterprise reality, dig into traceability and DFM, and explore how integrations unlock better simulation.

• Capturing design intent across CAD, PDM, PLM
• Linking requirements, tests, and design changes
• Closing the manufacturing feedback loop
• Reducing rework and documentation overhead
• Enabling simulation through shared context
• Funding path, pilots, and early customers
• Differentiation from text-to-CAD and new CAD software
• Partner strategy with incumbents and AI tools

Transcript

Welcome And Guest Intros

Roopinder 0:12

Hello and welcome to FODES, the Future of Design and Engineering Software Podcast. My name is Ripindertara. On the show, we will have guests that will discuss tools and technology that engineers will find interesting and useful. Today's guests are Kanal and Arjun Jane, twin brothers who have created Tandem, an AI-based knowledge layer that is meant to help design engineers create products. Arjun, Arjun.

SPEAKER_00 0:40

Canal should be joining in just a second. Oh, okay. All right. How are you? Not too bad. How are you doing? Good, good. Where are you located? So we're based out of San Francisco from the Boston area. Oh, are you in San Francisco as well?

Roopinder 0:53

I'm just north of San Francisco. I'm in Marin County.

SPEAKER_00 0:56

Okay, that's a beautiful area. One of our one of our friends is from uh Tiburon. Where in San Francisco are you? We're in the mission, just a little bit south, and then we have an office near Oracle Park area, so Mission Bay as well. Oh, okay. Okay. Hi Rupert, nice to meet you.

Roopinder 1:11

Hi, Canal. Nice to meet you. Are you guys okay? I know there's a lot of Janes. Are you guys brothers?

SPEAKER_00 1:17

Yeah, we are twins. Okay. Yeah, so we grew up in the same room, lots worked together at a lot of different places, ended up going to different schools, and then just started this company together with two other co-founders.

Roopinder 1:28

I was just reading about you, Canal. You went to Virginia Tech, right?

SPEAKER_00 1:32

Correct, yeah. I went to Virginia Tech.

Roopinder 1:34

And Argin, where did you go? I went to Purdue. Okay. All right. Do they play each other, those teams?

SPEAKER_00 1:39

We played each other once in football. So we both went to it was at Virginia Tech. My friends and I, we went down to visit Canal, we went to the game. It was a lot of fun.

Roopinder 1:47

It's amazing that twins grew up together and they're still remaining friends, actually. That's uh to your credit. I assume you're friends, are you?

SPEAKER_00 1:54

You have to be, right? We're friends, we're brothers, and now we work together as well. Okay, all right.

Twin Engineers’ Backgrounds

Roopinder 1:59

There are really two of you. This is not a split screen mirage, right? There are we may have to ask some proof of twinship. One of you worked at Boeing, correct?

SPEAKER_00 2:08

Yeah, so I worked at Boeing for a little bit doing commercial aerospace. I was on a team called New Features. New Features is a really small team. Like if you could imagine working at Boeing, it's not a super innovative focused kind of company, but I was lucky enough to be on a small team developing some of the new applications for the interiors of 787. So I was working with our high-end clients, if you think about like Emirates and the other airlines to develop seats, galleys, labs, mods, deco panels, etc. Working with suppliers as well as our internal design teams to create some of this, making sure that we are in compliance with the FAA, which is extremely important, especially during this era that Boeing is going through right now. And also Arjun experienced the same kind of thing at Rolls-Royce when he was working in defense.

Roopinder 2:51

Rolls Royce defense. So you guys have had both uh kind of like dream jobs, right?

Dream Jobs Vs Reality In Engineering

SPEAKER_00 2:56

Yeah, yeah. I was I was at Rolls Royce, I was on the F-35B fighter jet there at the time. I was a mechanical engineer, design engineer there based out of Indy, that's where Rolls Royce headquarters is. So I was there for some time. We were going from an older engine to a newer engine or lift system is what they call it. I was on the lift system. And during that project, one of my jobs was to really understand why we designed the system the way that we did 10 years ago. Like, why do we choose certain materials? Why did we choose certain tolerances? And was there like a specific reason behind those? And during my time, I started to realize how poorly mechanical engineers document their design decisions. And it made it super difficult for me to really understand exactly why the tolerance stacks were the way they were. And now my decisions were for the new design. It's like based on all this data, can we make accurate decisions to loosen tolerances, etc.? And because that type of knowledge wasn't there on the original decisions, we had to either make the decision to do more FEA analysis or just continue with the status quo. And because FEA and all of that takes a lot of time and money, we decided to just continue and not innovation innovate. So started talking with Canal, and there's got to be a better way to capture some of this type of knowledge.

Roopinder 4:10

It's a lot of I don't know if disillusionment is the right word, but you get your dream job as an engineer, you think you're gonna be really creative, you land a really good company, and you find out you're just looking for parts, you're looking for specs, you're looking for all this. I was dissuaded from that job at Boeing by seeing a room full of all guys in white shirts. Now, this was the member, this was 30 years ago, all guys in white shirts and ties on drafting boards. I thought, oh, you're just designing a screw or landing gear or something, not the whole plane, not what I fantasized an engineer would be.

SPEAKER_00 4:44

100%. Like in university, you learn all the theory, you maybe do a capstone, which takes you from design, inception, conception, and then actual designing and production. But in the real world, as a mechanical engineer, you start to realize a lot of your job is actually just trying to deal with bureaucracy and looking into stuff that other people had done and it wasn't done properly. And so there's not a lot of creativity and design work being done, and a lot of innovation is stifled as a reason of that. And then management's always worried about these three main things, which is like time, cost, and performance. And as a result of this lack of real understanding of what is being done in the design process, those three things often lack quite often and gets the projects off schedule or the performance of your team, the output of your team. You have to do a lot of part rework, et cetera, suffers. And in turn, your costs then go up with not knowing DFM, et cetera. So we set out to make the engineers be more creative as Canal and I always wanted to be, and allow them to actually do design work and innovation as clearly as possible to get there. And Canal and I have been designing ever since we were younger. Actually, growing up, our Christmas presents was like the new edition of SOLIDWORKS. So we've been designing for a very long time, and we want to bring that type of innovation back into the industry.

CAD Tools, Ecosystems, And Strategy

Roopinder 6:14

I often think that engineering is not hereditary because I had a lot of trouble passing my engineering jeans on. So I'm sure your parents are very proud of you both being engineers.

SPEAKER_00 6:27

We both have been using SOLIDWORKS when we were young. We're both CSWPs, have used actually a plethora of softwares, not just SolidWorks. Like we used OnShape at a couple of internships. I used NX at Rolls Royce, like Canal was using some Katiya. So we've been doing CAD out of various different softwares for a long time. And I've noticed a lot of them are similar, but they have their nuances, they have their differences. Some we like more than others. So you're creating something that will enhance. The way that we think about it is twofold. Actually, our goal as a company is, and we know how these large incumbents work and how DASO, Autodesk, PTC are quite closed off in a way, but we honestly feel like we are the type of company that actually can provide value and lock customers into a certain ecosystem, whether that's DASO, whether that's Autodesk, et cetera. And that's kind of partly why we're still trying to figure out like what is our path in terms of partnerships and what is the path forward from there. Uh, a key thing is we're not actually competitors with these incumbents in CAD softwares. There's a lot of AI companies, AI native companies who are trying to replace PDM, PLM, CAD, et cetera. And they're trying to create their own PDM, they're trying to create their own even CAD software, which Kinal and I have strong opinions is probably not the best thing to do because it's hard to get people to switch off of their current systems, right? Or companies are trying to do text-to-CAD and really help with specifically the design process and generating parts, even though they don't really know at an enterprise level, adoption of text-to-CAD is almost zero. Because if you ask a text-to-CAD model to generate a part for a turbine, it probably is not going to do the best job because it doesn't know what we believe are all the inputs and context necessary to do design generation, right? Which is requirements, manufacturing feedback within colleagues, conversational channels, teams, outlook, all of these types of things are necessary to actually do proactive engineering down the line. So that's really how we're structuring our company. How can we actually just first be a knowledge layer to capture all of the knowledge that goes into design, capture all of the design inputs necessary and see the relationship between parts? So we're not actually creating a new CAD software, a new PDM, PLM, but we're actually just an integration layer into kind of all these systems, as well as then integrating into the CAD software itself and helping with the design process. So we're not actually being competitors with a lot of the incumbents or trying to do text-to-CAD or design generation. We're first really trying to do knowledge retrieval and then eventually do the proactive engineering down the line.

Roopinder 9:19

Makes sense. Why not use CAD as a modeling engine, so to speak, and just improve on the maybe even the additional stuff that an engineer has to do that is not CAD. Leave CAD do the geometry. It's good at geometry, right?

SPEAKER_00 9:34

That's exactly right.

Why A Knowledge Layer Beats Text-To-CAD

Roopinder 9:36

Don't reinvent that wheel. It's it's pretty good at geometry. Not, I have to say, not all geometry, but it's good at prismatic geometry. There's still some said advances it could make to handle other types of, let's say, organic shapes that would be good, but uh but I digress. So when did this idea occur to you that hey, I should I should get AI assistance to help engineers do their jobs?

Origin Story And Team Formation

SPEAKER_00 9:59

Tell me how that really started for us. So when we were in college, Arjun and I have always been thinking about ideas in some format of starting our own company, et cetera. When we were in college, we were dabbling with this idea of how can we tokenize mechanical engineers' work. Um maybe you don't know a manufacturer, but you're a mechanical engineer that has a bunch of designs. How can you bring that to the market? Anyways, so we were dabbling with that. We started to realize when we were working on our full-time jobs that there's extreme opportunity in the industry for some kind of co-pilot or agents in mechanical design. And the reason why we saw that opportunity was we started to look at the landscape. And as mechanical engineers are interested in innovation, you start to like research on all the startups that are being created, and you find the ones that are really focusing on just specific simulation agents or just specific text-acad, et cetera. And as we alluded to earlier, was these tools aren't useful at the enterprise level. Maybe they're useful for makers or for hobbyists or people who have their own prototyping shops, et cetera. But at an enterprise level where you have thousands of parts integrating together, working together, you have context from the FAA requirements, FDA, you have manufacturing feedback, as Arjun mentioned. You can't really generate a part using Text Decader. You can't do small-scale simulation without understanding how all this context plays into each other. There's a gap in the market of what people are doing and what we believe is the best way to help mechanical engineers. Let's go try to tackle that with our knowledge in AI as well as engineering. I also ended up after Boeing working for AWS Amazon as a technical product manager there, helping to build some generative AI applications to help AWS solve more efficiently. So we had the know-how, as well as two of our best friends were doing machine learning research at one of them, was a software engineer at AWS SageMaker, so training foundation models. The other was doing some AI research at Google. So we had the team and the expertise in all the different areas. And I felt like we were the right people to tackle this problem. So that's how like how we came about it. I don't know, Arjun, if I missed anything there. Yeah, I just want to add like how we really came about the kind of core wedge and product we're doing today. We're eventually trying to do this proactive engineering. And because we've captured all this knowledge, we can do that. But what we really saw working at both Rolls-Royce and Boeing was there's two areas where engineers really lack. And that's what I was alluding to is like documentation and being able to have a knowledge graph, a knowledge log of the design process. And then, secondly, like requirement management and how that requirements actually link to CAD data and closing the feedback loop from manufacturing feedback, how that affects requirements, and how requirements affect design. And I was an engineer at Rolls Royce who was doing a lot of manual work with looking at requirements, looking at test reports, looking at test plans, and looking at the CAD and making sure all of them link up and then verifying and doing the verification process for requirements. And we were like, there has to be a better way to not only produce documentation for design sessions, like maybe I'm the past two weeks of work on SOLIDWORKS, I don't want to write a 10-page report to my manager or to my design review. Is there a system that can make that process much quicker? And then we started thinking like, what if we started watching people as they design? And that solved that issue, but that also in turn solved the requirements issue because now we can link exactly what people are doing within CAD design to the requirement that they're trying to achieve or to the manufacturing feedback that they're trying to achieve. So we're trying to close the kind of development cycle and make it more and known throughout the industry. So if I work on a part, Canal could open it and understand my process as well.

Roopinder 13:53

You guys are both well-educated engineers and you studied engineering. Did you also study programming and Python? Or did you have to pick that up? Did you program yourself or do you have developers?

SPEAKER_00 14:06

Yes, I think growing up nowadays, like everybody has to learn some form of programming. So in high school, we did take some Java classes as well as in college. However, we're not definitely the level of software engineering that's needed to build out a full-scale product. That's where Lohit and Location really are two co-founders come in, and we have this perfect blend where they're like amazing, super talented software engineers who can build extremely fast and who know how to use the most latest tools and how to integrate some of these AI applications in. So that's where the blend comes in. Arjun and I were dabbling in the beginning. We were able to create some MVPs and low hit and low case really are able to take our know-how in the mechanical industry and really bring it together to create our product. And that now we have some initial pilots.

Roopinder 14:53

You guys were the directors and you had producers.

Documentation, Requirements, And Traceability

SPEAKER_00 14:56

If we were to define it, that's how, but I think at the scale that we are, like and there's four people, everybody has to be doing everything. So they work with customers, they understand the mechanical engineering. Arjun and I help out with the software stuff. We're all doing everything, but to break it down, yeah, Arjun and I are on the product, business development, mechanical side, and they're on the software development, technical, AI side.

Roopinder 15:16

Your team initially did they have a concept ready to go? Did it exist on paper or did it actually exist in code? And you had to use that then to generate some funds.

SPEAKER_00 15:27

We raised a precede round in September by a firm called Antler. We had a prototype MVP type of demo, no customers at the time. We were doing some beta testing with some university, Virginia Tech Duke, etc. But so we we low hit and location, Kenal and I, we've been doing us four since the beginning. So we've all been working together. We're all co-founders. So we had an initial MVP that we were doing, raise some money. We all went full-time end of September, and we've been doing it since. And now we have two paying enterprise level customers, a med device company as well as a design consultancy firm. And we're trying to scale from there. We have a lot of cool customers in the pipeline that are willing to and want to test our do-it-and and pay for it. We're just trying to expand it and scale from now. So we have a pretty good product, it's not ever changing, of course, but that's where we are today.

Roopinder 16:21

I'm noticing this that people seem to be having startups seem to be having more success getting funding. I come from a time where you had to tell venture capitalists how to spell CAD. They had no idea. And now that seems, oh, you have a good idea. Not only do they know how to spell CAD, but they understand design and they understand a need for AI to be in design. So what is it was it not too difficult to get funding?

SPEAKER_00 16:49

I think it is difficult. I think if you have a good team of the people that are capable to do it, and if the VCs believe that you're capable, then of course it becomes easier at the pre-seed stage. I think they're also finding that now is a great time. If you see AI and and this really productivity has come to other industries before the industrial mechanical industry, software development has been extremely disrupted. You have tools like cursor, etc., doing extremely well, even cloud code, chat GPT, all these things. Then you have some of the consumer industry that's been really disrupted, again, chat GPT or Cloud, Anthropic, as well as some verticalized industries in there. And then now we're seeing, hey, why isn't this really happened for the mechanical world as well? And I think VCs are understanding that and realizing that this is important. I think the mechanical world is has its own difficult aspects to it. If we're able to crack the code, I think you can be extremely successful in the industry. We're planning to raise in the next couple months a seed ground to get an initial investment of not too much money, you need to show it's more about the team. And if you have a good team, then it's easy to raise. I think over some time it becomes more about the product and more about the traction, etc.

Roopinder 18:05

Okay. All right, well, good luck going before investors. We can talk about do you have any plans to also address simulation?

Funding, Pilots, And Early Customers

SPEAKER_00 18:12

Yeah, so I think today what we do with simulation is just through the test support for the medium MPLM. Just getting trying to collect that knowledge is what we're doing today. But we definitely see down the line as we're collecting this, we're creating this knowledge layer and helping the design process, there's a very close link with the simulation. We need to test your designs and to see if they'll fail under concentration, to see if the target is okay, etc. Because we're creating this knowledge layer and we had a deep understanding of how a company designs or what they're designing and the processor state that they are within the design. These other agents, these other companies that do AI simulation, that do text a CAD, they can plug into our databases and make their simulation agents and their text of agents ten times better because they have all the knowledge. For example, say we're working with the Boeing and we're helping them through the process that we're doing today, and then a simulation AI company comes in and Boeing says, Yeah, let's work together. Now they can make their software ten times better by using our database and our knowledge. So we're not doing simulation ourselves, but we can enable simulation to be done at an enterprise level for these AI simulation companies.

Roopinder 19:28

The two companies that you said were already you're already working with, are they your guinea pigs? Are they giving you are they giving you guidance?

SPEAKER_00 19:35

Probably, I think, especially in the first, I guess, a few months of our development. I think every company that we work with is definitely helping shape the product for sure. I would say, like, our true guinea things and the people that helped us in the very beginning, like understanding bug testing, etc., are these universities that we work with. I think students are an amazing asset. They work extremely hard, they develop amazing, and that they're open to trying out new things. So they were really what helped us get off, get our feet off the ground. And now when we have these more large-scale enterprise pilots, they're the ones that are really the balance and helping us take the end goal of what tandem is, but also we're providing a lot of value to them. So that's why we have that relationship that works out nicely. There it's a partnership between us.

Roopinder 20:14

Because at the beginning of the year, you'll be going out to whatever shark tanks and run out there.

SPEAKER_00 20:19

Yeah, I think closer to maybe March, February, March we'll be like talking to some investors and we'll see where we are in terms of fundraise. But yeah, I think around that time we'll have to do that.

Roopinder 20:29

Are you keeping your eye out for competition? There's like I said, there's a few companies we run across that are also now using implementing AI and engineering products or engineering software applications. Are you heard of others?

Simulation, Integrations, And Enablement

SPEAKER_00 20:44

Yeah, there's a bunch that we know of AI, yeah, I'm not sure. Yeah, collabation, Adam CAD, Simscale. There's a bunch of kind of people in the industry who are doing stuff that is similar, but we don't really quite believe is doing at scale the amount of integration that we want to do. And a lot of companies, like the ones that I mentioned, are really focused on helping specifically design generation. And we're more how can we help with document documentation and knowledge retrieval and kind of creating this knowledge rack, which that in turn will help with the design generation down the line. So we're taking a different approach to a lot of companies who are on on that side.

Roopinder 21:25

Yeah, the need to differentiate might become important as more and more players enter into this. Have you heard of quarter 20?

SPEAKER_00 21:32

Yeah, quarter 20. Yeah, yeah. True. We know them.

Roopinder 21:34

They make engineering wiki.

SPEAKER_00 21:36

A lot of these very cool there, they have some amazing technology. I think where we are a little different is we're really trying to capture design intent. The reason to why engineers are doing action their task. We believe that comes from context, requirements, manufacturing feedback, etc. So that's where our depreciation our where we depreciate ourselves is coming from this capture design intent, which is extremely important.

Competition And Differentiation

Roopinder 22:00

I was hearing a lot about the OAI. They're highly very visible choice. It's a big space right now. Engineers need a lot of help. And I'm so glad this is happening. I have admitted I struggle with my tools, like my CAD tools, like just picking them up, learning them. And I wonder why I can't have a natural language interface to whatever, to solid words or top shape, right? Or whatever. Anxys, right? Why do I have to become an ANSIS engineer to use ANSIS? I just want to be a mechanical engineer that uses ANSIS, right? So I wanted to tell it, hey, analyze this part. Here's the put the forces here, put the restraints here, it's made out of this material. I would just say it, just like I said to you, right? I I want that in my CAD program. I always use a bike frame as an example. Make me a bike frame. If the AI knows me, it knows that I make bike frames, it's got a bunch of bike frames already designed in my company, which I work for track, it knows all the track portfolio. It should then say, hey, this I want AI to be interactive. Like I want it to suggest things. Not just making a bike frame, suggest something. Say like a partner, be a partner, be that partner, be that assistant, right? Because I don't want to do that shit work, right? I don't want to I don't want to cope through all the hundreds of designs that are on file, right? I want to ask me, is it carbon fiber? Is it titanium? Okay, then here's the designs that I've made so far. How would you like me to improve them? What ChatGPT is doing nowadays. It asks me now, do you want to make a story out of this? It suggests space. I want my AI to do the same thing. I want it to be helpful. And understand, like you said, understand the context in which it's operating. And that context is what I make, what my company makes, also what the industry has of it. You guys know that, I'm sure you guys know about generative topology optimization. So there's a case of AI going bad. It starts making shapes. And this was the case of software developers say, I know how to make shapes better than you, dev engineers. And it would make these stupid shit that I had no way of making and no way of creating because it didn't use standard shapes. You could only 3D print them. It was absurd. And that AI waved backwards. It's not AI like we know it, but this was AI being provided to us and being us being told it. We don't think outside the box, but this software does. But the result was ridiculous because I'll use a bike frame example again. It didn't make a bike frame out of tubes. Tubular construction is a gold standard for bike frames because it's great at torsion, it's great in compression attention, and he would start making bike frames that look like blobs.

SPEAKER_00 24:33

We totally agree, and that's exactly why we started this. He saw again AI being thrown at design. And this time not really in topology optimization, but whether that's text to cat or a bunch of different applications, we felt like we still haven't understood what the inputs are, and those inputs are extremely important. Design for manufacturing. What tooling do I have available? What is the cost of the tooling? How much does it take to manufacture this certain shape? Like what are the specs and the best pages?

Roopinder 25:00

Yeah, so how about that? What are the specs there? What do I have to deal with?

SPEAKER_00 25:02

Don't make blobbies like our goal is how can we capture all of that knowledge? We call it design decisions. Yeah, how can we capture that and then eventually use this as parameters for amazing proactive engineering, partner agents, whatever we can call it. Yeah. That's exactly right.

The Assistant Engineers Actually Want

Roopinder 25:18

I use another example. I'll let you go. I know you guys have better things to talk to old journalists. The idea, like, okay, I want to fasten this to this, and it it should know. Hey, I use this kind of flags, these are the kind of bolts that I have available. This is where the holes should go. I don't if I'm making a bridge, I don't want to spend hundreds of hours designing the fixtures and specifying the hardware. It should just why can't it know that? This is something I would give an assistant to job to do. I expect AI now to be that assistant. So if that's what you guys are doing, I'd invest. Sure.

SPEAKER_00 25:51

Yeah, that's that's our whole uh that's our whole thesis of why we started this. Super, we're super excited that it resonates with you.

Roopinder 25:57

Keep me posted, keep me in the loop.

SPEAKER_00 25:59

Great conversation. When we ask some cool updates coming, we're happy to meet again and get on the podcast.

Roopinder 26:04

Uh, look forward to meeting you and seeing you in person. Very good. All right, thank you. See you. Bye-bye. Have a good day. Bye-bye. Thank you for listening to Faux, the future of design and engineering software show, brought to you by Ench Technica. I hope you have learned of a new application or technology that will help you with your job. If you have an application you think would be of interest to other engineers, please let me know by emailing me at repinder at enchtechnica.com or message me on LinkedIn.