I’m excited to share the very first episode of the Mission Matters podcast, a podcast from Shield Capital hosted by my colleague Akhil Iyer and myself that explores the intersection of technology, startups, and national security.
In our very first episode we speak with Alex, Ben, and Isaac, the founders of Distributed Spectrum, a young New York City based startup using ML to provide a comprehensive, real-time view of the radiofrequency (RF) environment.
Akhil and I decided to start this podcast because we really saw a gap in the podcast space: there are lots of podcasts about tech and startups and lots of podcasts about geopolitics and national security, but there are very few podcasts that cover the intersection of the fields. In this podcast we will really dig into the nitty gritty details of building and deploying cutting edge technology for national security customers. We’ll cover everything from technical implementation details (where to get data to train AI models? how to benchmark those models?) to acquisitions processes (what kinds of contracting mechanisms should startups prioritize?) to accreditation processes (what does it take to make it through the ATO process? what is needed to make a hardware system NDAA compliant?), hiring (how can startups hire good engineers with security clearances?), and more. Tune in each month for conversations with founders and government technologists about the opportunities and challenges for startups building cutting edge technology for national security customers.
In our very first episode with Distributed Spectrum we cover:
⚡️ The importance of electronic warfare on the modern battlefield
📈 How to train machine learning (ML) models that understand radio frequency (RF) data (hint: it involves synthetic data pipelines and building deep customer trust)
📊 How to benchmark ML models that understand RF data
🧑💻 How to hire a team able to build complex technical products for national security customers
And much more!
You can listen to the podcast on Spotify, YouTube, the Shield Capital website, or right here on Substack.
As always, please let us know your thoughts, and please let us know if you or anyone you know is building at the intersection of national security and commercial technology. Tune in next month for our next episode!
Here is the full episode transcript:
Maggie 00:04
Welcome to the Mission Matters Podcast, a podcast from Shield Capital where we explore the intersection of technology startups and national security. I'm Maggie Gray,
Akhil 00:14
And I'm Akhil Iyer
Maggie 00:15
And we are your hosts from the investment team at Shield Capital. Shield capital is a venture capital firm investing in early stage companies building at the intersection of national security and commercial technology. In this podcast, we discuss the technical challenges of developing and deploying commercial technology to national security customers, as told from the founders perspective. In this episode, we're joined by the founders of Distributed Spectrum, a young New York City based startup using machine learning to provide a comprehensive real time view of the radio frequency or RF environment.
Akhil 01:01
Today, the modern battlefield is more reliant than ever on digital systems, everything from drones to software defined radios to satellites and much more as such, modern service members are more vulnerable than ever before to electronic warfare, in which enemies adversaries are able to use electronic jammers to disrupt communication systems, disable navigation tools, identify positions and so much more.
Maggie 01:23
We’re super excited about what Distributed Spectrum is building. Current DoD systems cost millions and millions of dollars for just a few sensors and require months of training to use. In contrast, Distributed Spectrum systems cost just a few thousand dollars each require only a few minutes of training. In this conversation with the founders, we cover everything from accessing RF data in order to build effective ML models to actually interpret the RF spectrum, how to benchmark those models, the similarities between the tech needed to manage computer vision and radio frequency, how customers have actually responded to using Distributed Spectrums technology and much more. We're joined today by Alex Wolff, Isaac Struhl and Ben Harpe, the founders of Distributed Spectrum. Alex, Isaac, Ben, thank you guys so much for coming on the show.
Alex 02:13
Akil Maggie, thanks so much for having us on. I'm Alex Wulff I'm one of the co-founders and the CEO of Distributed Spectrum.
Ben 02:20
Thanks guys. I'm Ben Harpe. I am one of the co-founders and COO.
Isaac 02:23
Hey. I'm Isaac Struhl, and I'm the CTO.
Maggie 02:26
All right, so I want to start off with an easy question, Alex, Ben and Isaac, can you guys tell us the Distributed Spectrum origin story? Tell us a little about about yourselves. How did the three of you meet and how did three Harvard undergrads get involved in electronic warfare in the first place?
Alex 02:41
Yeah. So to answer your question about our origin story, so Ben, Isaac and myself have been friends since the very beginning of our freshman year at Harvard. Ben and Isaac were fresh in your roommates, and I'm pretty sure Isaac and I met on literally the first day of classes. We had spent a bunch of time all throughout college working on projects together, doing P sets, things like that. And right as we were about to enter our junior year, sorry, our senior year of college, the pandemic hit, and that really gave us an opportunity to, instead of probably graduating online and taking classes online, to say, hey, maybe the time is right for us to take your off from college and actually work on creating a company together, which had always been something we talked about doing. So the original idea behind Distributed Spectrum and what we started the company around came out of just my love of radio waves. I had been experimenting lots of hobbyist technologies for a pretty long time, and really fell in love with the concept of being able to utilize radio waves to communicate and also understand what's going on the radio spectrum. So I had done a few internships at companies like Raytheon, Lockheed Martin earlier on in college, and got exposed to just how kind of expensive but also very performant a lot of these existing electronic warfare and sensor platforms that our military relies on. So really, the core idea behind Distributed Spectrum was to augment some of these capabilities on the battlefield by instead of relying on a very small number of kind of large and expensive platforms, to instead use very performant commodity hardware that was starting to enter the market at the time, to change the paradigm to be to use instead a much larger number of lower cost systems powered by great software.
Isaac 04:22
Yeah. And so the thing to understand is Alex was the radio guru. He had just come off of writing a textbook while he was a junior in college on beginning radio theory, but Ben and I knew that we could use our backgrounds in computer science and statistics and machine learning to solve this really interesting problem, and that's what drew us to it at the beginning, because it was just something that was really interesting to solve awesome.
Maggie 04:42
Alex, could you tell us a little bit about what actually is electronic warfare? I think this is something your average person probably has never heard of, or at least has not heard of until recently, and what makes electronic warfare so relevant today?
Alex 04:57
Yeah, definitely. So we view electronic warfare. As understanding and manipulating the radio spectrum to your advantage outside of the oceans, pretty much anything that's communicating wirelessly is using radio waves to do it, and it's no secret that radio is becoming increasingly more important on the battlefield every day due to the proliferation of unmanned systems. So this presents both a huge opportunity to understand what the adversary is doing based on all these signals and also a threat in that now our own signatures are becoming increasingly greater that the adversary can exploit. So electronic warfare is just growing in importance every day as more and more devices out there starting to use radio waves in order to fulfill their functions.
Akhil 05:38
Isaac, this stuff isn't new, right? Electronic warfare goes back certainly before WWII. But I think in the modern sense of the electromagnetic spectrum, from radio waves to radios, is really WWII on. This stuff is not new yet. The idea of sensing the environment electromagnetically, figuring out how to either decept, deceive, counter and address it is a new so what's what's different today, particularly as you all saw this coming into a space where, again, there had been an extensive literature of use cases, of applications, of technologies. What did you see that was missing here?
Isaac 06:19
Yeah, so there are a couple things. The first thing is that there are so many more devices on the battlefield that are using the wireless radio spectrum to communicate. These devices also might be at low power, and they also might be on for a very short period of time. And so if you want to just sense these devices effectively, you actually need to be close and you need to have sensors everywhere, which means you need lots of devices. And what we've seen in the past is, you know, a small number of very large, expensive devices that required lots of trained operators to effectively get outputs from them. The second thing is that just the concept of operations during the global war on terror looked very, very different. We could fly a plane with a dozen of these trade experts to focus on a single target at a time, and we can't do that anymore because of the sheer number of devices emitting RF. The fundamental way that we that we do sensing, needs to change in the great power competition context.
Ben 07:05
Yeah. And just to expand on the scale of the problem that Isaac is talking about, because I think this is really what has changed. You know, first is the geographic scale. You want to be able to cover the entire Pacific Ocean, there are going to be anomalies popping up, weird things that you're not expecting, the spectrum that you want to know about, and it doesn't make sense to station a large number of really expensive sensors all across Pacific. You just can't really do that. The other thing that's changed is really the skill of the missions. No matter what you're doing, you're going to be affected by the radio spectrum, and so there is this really strong need for anybody, regardless of their mission, regardless of their amount of training, to be able to inform their mission by what's going on in the spectrum, and so that's really the difference that we see now, is that not only do we need sensors everywhere, but everybody needs to be able to understand the outputs of those sensors and how it impacts their mission.
Akhil 07:52
Thanks Ben, that's really helpful, and I think highlights the focus you all have on proliferated systems. How do we get every individual to be a sensor in this low cost, disruptive way to enable broad awareness across electronic support as you think about that and where you have come since the Harvard College dorm rooms to where you are now, can you talk a little bit about how the customer side found this, championed this, and recognize the same vision that you had about the future of electronic support and RF sensing generally in future environments.
Alex 08:29
Yeah, definitely. So pretty early on our one of the focuses that we had was just sitting down and talking to as many customers across the Department of Defense as possible. And that's one thing that was really interesting to us was customers are super willing to sit down, even with a bunch of kids right out of college, and just tell them about some of the problems that they face on a day to day basis. And we heard a lot of really interesting problems that people were facing with regard to the radio spectrum that we actually felt like we could tackle. Honestly, there was a lot of low hanging fruit out there that we heard about simple things like, hey, if my radio stops working, I don't know if it's because of some electrical or mechanical failure, or I'm being jammed, or maybe there's, you know, one person in my unit that can actually operate this expensive signals intelligence gear. But if we don't have them, then we're completely blind in the radio spectrum. So that really became our focus early on, was building capabilities to tackle some of these seemingly simple problems that are, you know, in hindsight, a lot harder to actually solve, but allowed us to give people just a lot of leverage by expanding their ability to conduct their missions, by giving them pretty tactical level situational awareness about things going on in the radio spectrum.
Ben 09:35
That started a really good cycle for us as well, where we were able to learn about these first problems show up and actually provide demonstrations and show that we could, with our pretty inexpensive hardware and with our software, provide much better solutions to these mission sets, and then from there, being able to expand our relationships, meet new potential customers, learn about their problems, and then continue developing our product to be able to tackle a much larger. Set of more complex missions.
Maggie 10:01
So it sounds like there really has been this revolution in the technology with these cheaper, software defined radios, better edge deployable ML systems that you're able to develop. I was wondering, you know, could you tell us about the first time that you really felt like a customer could use your product? You know, first time you saw them and you felt like they loved what you had built for them.
Isaac 10:23
Yeah. So, as Alex mentioned, one of the things that the DOD does really well is let new people understand their problems and develop solutions for them. So we were basically kids out of school, and we were able to embed directly with the customers and see their missions. So one of the first missions that we did was a full mission profile exercise. So we went down and we had a few of our inexpensive sensors scattered around the area, and these customers told us nothing. We didn't know what they were doing, and we certainly didn't know anything about the RF spectrum that they were using. So during that exercise, we were able to use our system to understand the spectrum broadly, but also show these guys what they looked like and what they were doing in real time. And in fact, one of the things we did was we did was we said text messages, text message alerts for them as they were moving around the exercise. And so the reaction was, Oh, wow. Like, this is really possible. You don't have to use super expensive hardware, and you can do it really, really quickly in, you know, real time or near real time. And so the two things that gave this aha moment, or that it was technically feasible, you know, you could actually do this with the systems, but also that we can make the output usable to somebody with no training. Everyone can get an alert and do some action with it. Yeah, exactly.
Ben 11:27
And one thing that we really try and focus on is we're not trying to replace electronic warfare officers and SIGINTers. We believe we can help these people do their jobs a lot better, get much more information more quickly. But then there are all these users who don't have any of that specialty training. And so one fundamental thing that we've just seen a lot is that people need to be able to make decisions on their missions based on data that they just don't have. And so another example is we were deploying sensors with ground based operators, and the training that we gave them was, just take one of our sensors, turn it on, plug it into your ATAK device, and that's it. You're able to start seeing alerts for the types of devices that you care about in real time. And that was also really cool to see just being able to run really minimal training, and seeing users who usually don't get any sort of spectrum information, having that seamlessly incorporated into their missions.
Isaac 12:15
The final thing is, just this simple understanding is something that's been core to us at DS throughout the entire time we've been building we think that everyone needs to understand the RF spectrum for any moderate mission, and so we just want to democratize that information.
Akhil 12:25
Alex, it sounds like there's potentially a lot of users of this, but you kind of have to prioritize, as a young startup, where to focus that effort. How do you think about that?
Alex 12:37
So the field of electronic warfare is obviously pretty broad, and very early on in our company, we spent a lot of time chasing, I would say, lots of different ideas, and we pretty quickly realized that we needed to focus around a few core capabilities that we could actually start bringing to market. A big part of this was deciding to say, hey, we need to build a few prototypes, and we need to think about exactly what missions we're going to try and solve. And those missions that we started to focus on were really around two core central concepts. The first is providing immediate awareness of threats in the radio spectrum to individuals that don't ordinarily have access to it. A big part of rethinking how we fight in modern conflicts is giving any warfighter, regardless of their level of training, awareness of both their own signature, but also immediate awareness of any adversarial activity, and that's one of the kind of key concepts that we decide to focus on. The second was, you know, realizing that we need to empower those that actually do have signals electronic warfare training to do their job a lot more effectively. There's not enough of these individuals out there, and there's a lot of tools that we can actually provide to them to make them be able to do their jobs more effectively. Think things like being able to throw out the 99% of signals out there that they might not care about and allow them to focus in on the 1% of things that actually matter. So that kind of key realization early on, of prioritizing around a few core mission sets was one of the things that really kind of helped us get going?
Ben 14:01
Yeah, exactly. And I think at the beginning, it was really about trying to find those missions where we could really demonstrate that our technical approach works, that we can build trust, that we can get more embedded in, you know, the networks of users who are going to care most about our product. And then as we grow and we have more under our belt, it becomes a lot more focused on long term, more strategic growth in the company. So starting to think about, really, where is the money coming from? Where are the big strategic what are the biggest strategic priorities that are coming and, you know, we think that we can have a pretty general approach to EW and to spectrum awareness that supports a huge variety of missions. But within that, we're at the point now where we have to really start going in the order of where the biggest opportunities going to be for the business.
Isaac 14:39
The final thing I'll add is that there's this marriage that we continuously try to make between business outcomes and product outcomes. So Ben and Alex just talked briefly about the business ones, but from the product perspective, we want to start by solving really, really specific problems for individual warfighters. So for example, being able to put our sensors in Ukraine gives us real, real world data on both. Of what the RF environment looks like, but also how people are actually using these systems to understand the situation around them, and what data is actually useful to them. And that type of feedback lets us iterate really quickly on the product, which ties into our business development activities as well.
Maggie 15:13
I wanted to pull on a few threads there. Ben, you know, you mentioned the importance of gaining trust, the importance of getting visibility the importance of really having a big impact on your customers. You know, one of the things that I found interesting about you all is that you've decided to focus, at least initially, on really the low cost and unclassified work. And I know historically, a lot of electronic warfare development has been pretty secretive, and I know it's something that a lot of startups working with the DoD have to navigate of trying to decide whether to go after classified work, whether to do unclassified work. Obviously, classified work requires hiring people with security clearances, going through all kinds of complex certification processes and much more. So can you tell us a little bit about why you all have decided to stay in the unclassified space so far, and any potential challenges that you have encountered with dealing with these classification issues?
Ben 16:10
Of course. So the first thing I'll say is that at the beginning, we didn't really have a choice. We were starting from school. We didn't have access to classified data or to classified requirements, and so we were forced to focus on how we can solve these problems from an unclassified level, and that's definitely started to change a bit. Now we're going to be able to tie into classified systems and take advantage of classified information. But it was actually really helpful to be forced to think about this from an unclassed level. There's actually a huge amount of value that you can unlock, filtering through, cutting out all the noise in an environment, alerts about the types of transmitters that are in your environment recognizing anomalies that we've never seen before, and it's a very different approach from the old way of having a static, classified signal library that you're directly comparing against. And so it actually means that we can be really flexible to how quickly things are changing the environment and giving people alerts and understandable insights about what's actually important to them.
Alex 17:00
Another point to add is, one of the other things that we're focused on is being able to build systems that we can distribute to partner forces, and in most cases, those do need to operate at the unclass level. So in that case, that actually gives us an advantage. And the last thing I'll add on is, for example, if we are putting sensors out in the environment and leaving them unattended, not having anything classified on them means there's no risk of losing that information, which gives us a lot more flexibility in our employment of our systems.
Akhil 17:26
Curious also, to ask you all are based in New York, not my immediate thought for having a company based there working in this national security space, but I'd imagine there's some actual benefits when it comes to being in New York.
Alex 17:38
So one anecdote that we tell our customers all the time is that New York is actually probably one of the densest signals environments on the planet. We spend a lot of time just walking around the streets of Manhattan with sensors in our backpacks. And one of the really interesting parts about New York is that, you know, on a nice summer day, on any given street corner, there could be 1000s of devices all broadcasting. So being able to have that as our home environment really gives us a lot of confidence that our systems are going to work in very, very congested environments, which is not something that we can just take for granted. One of the other awesome parts about New York is people want to live in New York, so it's been nice to be able to attract awesome engineering talent to live and work in the New York City area.
Isaac 18:16
What's really funny is we often end up going to exercises that try to simulate urban environments or in other urban areas. And those urban environments don't compare to New York at all. Like New York is so, so much more dense with respect to the devices that are out there. And so we'll get to these exercises and be like, Oh, look, a couple dots on the screen. Like, that's all I have to think about. It's just totally, totally different from looking at the outputs of our tests in Central Park or your office. And so the key point is, like when we do this, we have a much cleaner slate to work with, and that makes the performance a lot easier to achieve.
Akhil 18:48
Thanks Isaac, I'd love to dive into this, this concept of ease of use, and the parallel I'll make for those who are technical, is where the computer vision space has come right. 10 years ago, it was you doing a lot of the work yourself. Nowadays, the litany of tools, whether that's you only live once models, has really just proliferated, and the DOD has been a leader when it comes to lowering the barrier of computer vision access for operators. And the joint Artificial Intelligence Center and the work around project Maven is an example of that. What I sense, maybe, is that the way in which the computer vision world, both for national security and commercial, evolved into something that was easier to use, that that might be coming when it comes to the sensing of the electromagnetic spectrum. So my question as Isaac is, is that right? Is that how to view this, this evolution in radio frequency sensing, RF, EW, generally, and where will it be in five years?
Isaac 19:42
Yeah, it's a good analogy. And actually, I think it's even more true for electronic warfare. There are two main factors. I think the first is that the average soldier on the battlefield really needs to have a lot more information than they currently do about what's going on in the spectrum. That's just table stakes for being able to operate effectively in a congested environment. Written against a near peer adversary. They need to know things like if they're being jammed, and they need to know if something's threatening them in their area. The second thing is that we need to enable especially trained people that already have a lot of knowledge of signals to do their job effectively. We think there's always going to need be a need for these people, but now there's too much information for them to process, you know, by themselves, and so software and AI can help them understand things like, Hey, this is an important signal. This is an important anomaly. Fundamentally, this is something to focus your training on and focus your analysis on. With computer vision, I think we saw something similar. There's too much information. AI techniques can get actionable insight out quickly to the people who need them. And so with CV we ask the question, hey, if we can process camera data at the edge, why can't we put cameras everywhere and get their outputs automatically. The same thing is now happening with RF, because we're able to process RF at the edge, and therefore we can say, what happens when we put RF sensors everywhere and make the models good enough to run at the edge. And the key point is that it's possible now, because we can make these edge models good enough, and you can buy a software defined radio for like 30 bucks, I think the key difference is that it is much, much harder to interpret the raw outputs of an RF sensor than it is for a camera. So having software that can do that really difficult interpretation at the edge and just give back human level understanding is much more important for RF than it is for CV. So that's why I think this change is even more of a step change than it was for computer vision.
Maggie 21:19
Yeah, so I'd like to shift gears a little bit and really dive into some questions that I have about the actual technology that you guys are all building. And some of this is related to the question of classification, you know, other kinds of machine learning tasks, like computer vision, which we've talked about in the past, have a lot of easily available open source models. There's also a lot of open source data sets. I mean, obviously all of YouTube is image data. There are all kinds of models, you know, YOLO v3 and others in the computer vision realm. There are also low code, no code, platforms that make it really easy for people to customize other kinds of models. However, at least my assumption is that there are not large open source RF data sets available on GitHub or hugging face for you guys to just train some ML models. And again, I have to imagine there aren't just off the shelf models you guys could easily download and plug in. So I'd be curious to hear, how have you guys gone about training these models. Where have you been able to actually find good data to train them? Are using synthetic data? Is it about customer partnerships? Could you talk a little bit about that?
Ben 22:30
Yeah, you're exactly right. There are no off the shelf models for RF, and so a lot of what we're doing is creating our own custom architectures to be able to understand specifically the RF spectrum. We of course, take advantage of a lot of the research and really successful ideas from modern machine learning to inform how we do this, but we also have that added constraint of having to run inference in real time on our edge sensors. And so there are a lot of challenges and constraints that come from that that mean we really have to be building our own foundational models from scratch. And to your point, the other thing that makes this hard is that it's not like there's some huge existing data set of RF data that we can just download and train against, but an advantage that we have is that, unlike something like computer vision, where it took a lot of research to be able to generate realistic, artificial images, RF is a much more structured problem. You know, we know a lot about the physics that can train these signals, how people might want to modulate and code data in signals. And so we've made a big investment in really good hardware in the loop systems to be able to generate and collect a large amount of very realistic RF data. We can stimulate different combinations of how different signals might interact. We can simulate different noise patterns. And then we can also mix in real data that we've gotten from customers, from exercise, from CAST runs that we've done, and so we're able to have a large but realistic and flexible set of data that we're able to train on. And I think that's what's made us be able to be successful.
Maggie 23:51
Isaac, I wanted to return to a point that you made earlier. So you all have to deploy these models on Compute edge, constrained hardware. You know, there's no AWS or Azure out in the field somewhere. The idea, right is that these are things that a soldier could carry, that you could put on a drone. So I'd be curious to hear, how do you all think about the trade off between performance and model efficiency when building these models? And a related question, how do you actually test and benchmark your model performance? Again, I have to imagine there are no excellent open source benchmarks for you all to work with.
Isaac 24:27
That’s exactly right, so sure. So starting with the compute constraints, again, we're designing this with a particular mission in mind. So the first question is, what hardware can we get away with? Can we use a GPU? Do we have the power available for that? Do we have the size, or can we explicitly not use a GPU? Do we have to specifically use what the customer already has? Or maybe, can we use a server class, GPU? Like all of these questions go into it, and that means that our deployments often end up involving essentially unique hardware. So that requirement is both a blessing and a curse. The blessing is that we can make our approach specific to the two. Deployment. So when we have more compute, we can run bigger models, and when we have less we can do more specific optimizations. The curse is obviously that if we don't structure our model development properly, we might have to build a single custom approach from scratch for every deployment, and we certainly don't want to do that. So I think the key in answering this question is a modular approach, and also developing a, like, really deep understanding of specifically what the customer is trying to do. I think there's like, two main examples of what a customer tries to do. The first example, in you know, ML terms will basically be, I want no false positives. So for example, I want to use this model to tell me when to make a big decision like change my comms plan. And so I want to be really, really sure that I'm seeing the signal. The second thing is, I want no false negatives. So for example, I want to use this model to just triage important information for an analyst downstream. So make sure that we include everything that might be my signal. The key point is that these examples actually let us run much smaller components, because the tasks are a lot more specific, and that makes it easier for us to be flexible to different compute environments. And so the key point is, how do we actually get to this level of specificity, which lets us be flexible, and that is a function of how well we can create these real benchmarks for performance. So we have some lab benchmarks for some of our middle for our bigger models and for the smaller ones that I was talking about. We work really closely with the customers to figure out exactly what they care about and what they don't, so that we can really optimize the performance.
Maggie 26:26
One more related question I know, you guys are not deploying this technology in a vacuum. A lot of your customers have already purchased, you know, maybe millions of dollars worth of legacy hardware that you might actually need to integrate with. Can you share your experience? You know, challenges, surprises of integrating your software with existing DoD hardware vendors?
Alex 26:48
So of course, there's a lot of equipment that's fielded on the battlefield right now that is both very sensitive and performant, but also very expensive, and this equipment is primarily designed to be operated by a trained human at the same time, there is also inexpensive, commercial off the shelf hardware that's primarily designed to be operated by software. So a lot of the work that we have to do with existing department of defense vendors that we collaborate with is work with them to actually build hooks into their systems to allow our software to control it, as opposed to just natively having those hooks for commodity, commercial, off the shelf equipment. As an example, think about a system that's designed to detect and intercept communication signals. Instead of a human slowly tuning that radio at the speed of them being able to understand the information coming out of it, we want to be able to quickly tune that radio at the speed that our software can identify new things popping up in the environment that are interesting. So that's just one of the examples of the ways that we work with some of these existing vendors.
Ben 27:46
Yeah, and of course, we recognize pretty early that a lot of the customers we want to work with have already bought a lot of systems. And so from you know, at the beginning, it's not a great sales pitch to just go in and say, well, you should throw out all of that stuff and just use us. Use us instead. And the scale of the problem is so big, and the systems that are out there are already actually generating a lot of really useful and really important data. So I think the way that we think about this is less as US replacing these systems, but really from the software side, we can add automation, add filtering, add alerting, and really let customers deal with a huge amount of data that's either coming from us or coming from existing systems that they already have.
Akhil 28:23
Thanks for that. Ben, do I have it right then that in a lot of ways, what you are doing from this low end, cheap, commercial software defined is a complement to a lot of the existing systems.
Ben 28:33
Absolutely, that is exactly how we think about it. It just does not make sense to try to replace all the existing hardware. Sometimes all you need is just a software integration. Or maybe what you need is a small edge GPU that's co located with the system and processing its data. Or maybe it's a network of inexpensive sensors that then triggers one really exquisite sensor to go get more high fidelity information. We're also not developing any of the hardware ourselves, the SDRs, the GPUs, so we're going to be able to take advantage of improvements that other people are making to that hardware, and we're also expecting that a lot of these devices are going to end up on the field anyways. So really, the most important thing for us is to enable that the data coming out of all these sensors, regardless of their cost or their size or how they're deployed, we want to make sure that that data is really useful, and we want to make sure that it's understandable to anybody who needs it.
Akhil 29:20
Thanks. Ben, Alex, you talked a little bit about new, novel techniques coming to play here. And I don't think you can find a weekly economist edition that does not, in some way either, every other week talk about the challenges, really, with the pace of change, the pace of evolution, not just in RF and Ew, but across the modern battlefield, it reminds us, Maggie and me here at Shield, investing in the cybersecurity domain, a domain where inherently, there's a ton of new energy and new startups, because there is an adversary. There's an adversary that is constantly changing, improving, manipulating, influencing and shaping what the. X vulnerabilities are as you think about how you mature and stay ahead of that and stay at pace with the evolution happening, whether that's now, whether that's in the future, how do you think about that, both in terms of the software you're trying to deploy, as well as the underlying hardware, which, as you know better than I do, is increasingly constrained because of some of our supply chain issues.
Alex 30:21
Yeah, you're bringing up a very interesting shift that's happening right now, and I think it speaks to a few separate points. So the adversary exploiting vulnerabilities in a system that you deploy and adapting their techniques is pretty much inevitable, and we're seeing that right now on the ground in Ukraine. If you ship an electronic warfare system to Ukraine with a fixed and pretty brittle library of signals, it's going to become obsolete in literally a matter of days, which is something, again, we're witnessing right now. So any software that you end up shipping is going to need to have some ability to adapt and change based on the environment as the environment adapts and change. And that's one thing that we've been focused on is being able to make a system that not only can understand what's going on, but also have some ability to respond to changes in the environment. And the second piece of it is, you know, we know we're not going to be able to get everything perfect the first time, so having some ability to actually deliver updates to the software the systems that have already been fielded is also pretty critical, and it's also been something that we've been focused on. The second piece here that I also want to highlight is just the ability of making a system that's incredibly easy to use. If we can give people a flexible set of tools that they can use to accomplish their mission more effectively, it's pretty hard to circumvent human ingenuity. So one of the things that allows us and our customers to be successful is giving them access to systems that can just do very, very simple tasks and then do them well, to allow them to adapt to conditions as they change.
Maggie 31:48
So it really sounds like you all are going after a really complicated problem that you're trying to solve. It requires deep domain expertise in software and hardware in government, go to market. How have you all thought about building out your team to have different areas of expertise in order to make Distributed Spectrum a success?
Ben 32:11
It has definitely been a challenging team to hire for. As you said, there are a lot of very particular skill sets that we're looking for that really are not very common. So, I mean, there is a pretty small overlap between the people who are, like, really deep RF signals experts and people who are really strong ML engineers. And so when we started, the main focus was really about setting the technical back, one for the company, for the things that we really thought were going to be important going forward. You know, we have really scalable and repeatable access to data. We have a really strong and reliable embedded system that's running on all of our sensors, we can make a lot of progress on fundamental research on just generally, how to use AI to understand the spectrum. And so being able to find engineers who have been really critical to that success that we've had so far, we knew that it was impossible to find people with every single precise piece of experience that would be relevant to us, but we were able to find people who really understood the vision, who had enough experience to be really flexible to a wide number of technical ideas, while still being able to operate in an environment with a ton of uncertainty. And I think that changes a lot more as the company scales. When there were five or six of us, everybody has to do a bit of everything, and that's still very much true right now, where we are at 14 people and beyond, but we do have a lot more freedom to be able to add new people with very particular pieces of experience that we know we need to solve particular problems that we know about, to work on this particular mission with a particular custom customer that is, you know, I think both how we're thinking about scaling the team in terms of, you know, keeping that generalist mindset of being able to be very flexible, but also focusing on the specific value that we know We need to add to customers.
Isaac 33:41
I completely agree with that. I think the main thing I'll add with respect to Team construction is that the difficulty of this problem and the interestingness of the solution is what has really led to what our engineering team looks like. So right now, we're vast majority engineers, and I expect that we'll stay that way for the foreseeable future, just because the things that working on really like fundamental problems in signal processing and machine learning and stuff like that. And so, you know, we really need to make sure that we are solving these in a novel way. And so that's like, been informing how we make these hires on all of our engineering teams.
Akhil 34:10
Guys, you've been at this a couple years, made some great progress, both on the technical side and getting users really involved and championing some of the work you have. What's been the biggest surprise?
Alex 34:21
So, yeah, I talked about this before, but I think, you know, one of the most surprising things to us, especially early on, was just the willingness of end users in the Department of Defense to sit down with us and share some of their problems and capability gaps. And these early touch points were so critical early on in shaping our thinking about how electronic warfare should be actually conducted. So I just think there's something so valuable about providing environments for individuals with no preconceived notions or prior biases about the way that things should be done, to speak directly to end users in a setting that's free of filtering and free of basically things that just reduce the quality of that information. And it goes deeper. And just having conversations too, but, you know, providing access to military bases and training exercises to see how some of these things are actually conducted on a day to day basis. And this is just, you know, incredibly critical to fostering innovation, developing novel solutions is something that the Department of Defense needs to keep encouraging.
Isaac 35:16
Yeah, and I think this is changing now, but when we were at school, nobody was doing it. I mean, when we talked about solving problems for the DoD or the intelligence community, the reaction we universally heard was, Are you crazy? That's impossible. Go do something else. You know, you won't be able to get any information. It's impossible to sell. Nobody will talk to you. And as Alex mentioned, we really haven't found that to be true at all. And yeah, like, one of the best parts about our jobs is that we get to go to customers and see their tech and see how they operate, and we get to be the ones to help make it better. I mean, it's awesome. I think this is, this is clearly changing. Now we've seen, you know, lots of defense startups and successful offices like the diu and others actually proving to us that the DoD is open for business. But, yeah, when we started, we got tons of advice about how this would be impossible. And then so far, we've seen this not and that we can build a product effective. Product effectively in this space.
Akhil 36:04
That's awesome to hear Isaac, and great to hear, particularly from the types of customers willing to champion and willing to accelerate your technology forward. I'm sure, though, that there are still some challenges. What's the one that stands out to you from the time you were in the dorm room together in Boston to now?
Isaac 36:19
Yeah, it's funny. It's funny. It's actually something that's pretty similar to my previous answer. But really, just getting the baseline knowledge to be able to speak DOD and communicate with people is something that we feel like we're continuing to evolve, but was actually really, really hard to do. And you know, when you first start, you hear tons of acronyms and you look them up and you start to vaguely understand them, but it's really more about the unique perspectives and the deep training that everyone gets throughout their entire careers at the DoD that leads them to think the way that they do and make decisions the way that they do. And very often, there are really good reasons for everything that's being done, and it's really hard to understand that context when you're coming into a new conversation. So for example, you know, the sheer number of program offices and people and jobs and roles and things like that is something that's hard to get rate not because you can't look up what their function is online, or look up what they're working on, because it's actually hard to understand exactly what led them to this point and how the culture knowledge and the system works influences the outcomes. And that's something that we've invested obviously, a lot of time in learning, and something that happens just as you do this more and more, but it's still very much a barrier and something that I think is starting to be broken down, but it's still definitely a challenge for us.
Alex 37:20
One challenge that I'll add on to that is, you know, we talked about how people are willing to talk to you, but going from the step of actually finding somebody who's willing to talk to you to put you on contract is a completely different story. And that was definitely, you know, a roadblock for us early on, and it's been a roadblock for lots of young companies, is being able to cross that first milestone of actually getting your first paying customer and getting your first, you know, paid system out the door, and that is, you know, that's obviously something that comes with the time, but once you get there, the road gets a little bit easier. And, you know, demonstrating that past performance is important.
Akhil 37:53
That's awesome. Alex, yeah, aligning the user with the buyer and the one with the wallet is huge. And you guys have navigated that really well over the past couple years, and look forward to seeing what's to come. Ben, last question to you, what's been most rewarding?
Ben 38:08
Yeah, we talked about this earlier, but it's really just being able to feel like we're actually having a strong understanding of the mission need and that we're actually having an impact on. So you know, those moments that we were talking about earlier around seeing people use and get value out of the system that we've built is really exciting and really rewarding. The other thing that I think is personally particularly rewarding for me is that I think I have pretty strong technical opinions about, you know, what is a good way to solve these problems in the RF spectrum, how we should be approaching these problems. And so one thing that's been fun is we're often met with a lot of skepticism around, well, this is the way I've been doing RF for 40 years. Why would we change it? Or, you know, I don't trust AI, because it's going to hallucinate. I don't want that to be involved in how I understand the spectrum. And so with this strong understanding of the mission, being able to show very concrete examples of, okay, well, for this thing that is really important for you to do using a software defined system, using machine learning, lets you achieve this outcome that was previously impossible. And so showing people that and seeing their response is really rewarding. Like, you know, you can just see, okay, I know this is going to be useful. This is going to actually have a difference on people's missions. And I think that justifies, not only the ideas we've had, but, you know, the way the work that we've actually done.
Maggie 39:16
It's awesome to hear that you are really able to see the value of what you're building in customers hands. So our very last question for you guys to close out this episode, what advice do you all have for other founders looking to build a startup in the national security space?
Alex 39:33
So one point that I would start with is one of the key prerequisites to having people kind of willing to take a chance on you, willing to trust that you can actually solve their problems is having something physical to show them, until we had a real prototype that we could actually put in front of people and have them sit down and actually use it. It was actually pretty slow going for us, and we didn't necessarily make as much progress as we wanted to, simply because we had all these ideas in our head, but we had never actually spent the time to sit down and go through and say. So okay, to actually solve this person's mission, here's the product that we need to build, and let's just get a rough prototype out the door to actually do that. And I think we were just a bit overwhelmed by the total amount of things that we could actually do. So having something physical to put in front of a customer is just the first thing to show them and iterate on is extremely important, and it's something that's just extremely necessary to doing business with the Department of Defense.
Ben 40:22
One other thing that's really stuck out to me is just the value of doing things in person. In general, I think that pretty much, no matter what you're doing if you're building a startup, if you're trying to learn about your customers, I think that having in person interactions is so important. Our entire team is five days a week in our office in New York, and that's been really critical for us. But the real mission learning and particularly building trust. I really think that only happens in person, and particularly in the national security space. The value of just meeting people in person, sitting down with them, really spending time to get to know them, you just can't do any of that over zoom Carl over a teams call. And so really, I'd say for getting started, it is really not only about just talking to as many people as possible, but trying to do as much of it in person as you can.
Isaac 40:59
One final thing I'll say is that I think there's a tendency, and certainly I feel this in myself. A lot of you know, young founders who see a system that clearly has some problems and say, Well, I can just rip this apart, and I have a much better solution, and people are going to love it, and this is the way to do it. And then, you know, they want to be visionary leaders. And I, I totally think there is a time and a place for that, but I think one of the most useful things that we've done is try to deeply understand, for example, why we're hearing some pushback, or why we're hearing some frustration, or why we're hearing about how a system works in a certain way, even if we think that system itself is crazy. You know, it's no secret that lots of that the DOD has lots of bureaucratic challenges, and that contributes to how to do, you know, sales to DOD. But there's a reason that a lot of these systems got built this way, and there's a reason that a lot of the programs exist the way that they do. So, you know, being able to adapt and fit into those, even if your goal is to ultimately make something much, much better than that, is something that you really need to be able to understand to get your company off the ground. So I would say, you know, pretty much everyone that we've met has great intentions and has clearly defined roles and jobs. And I don't think we've met really, anybody in the DoD who's been out to screw us right, like they've been really honest about what they want and what they want to create, and so being able to listen to that honestly and be able to help out with their with their missions in an honest way has been really, really useful to us.
Akhil 42:09
Isaac, I love that part, and I think really aligns with what Maggie and me work on a daily basis, and that's technology that has a an important mission, a mission that matters. Ben. Isaac, Alex, awesome, chatting with you all today, excited for what this year has to offer in terms of building and scaling this idea and this vision around the future of electromagnetic sensing and RF sensing broadly. Really look forward to seeing what's to come.
Share this post