Jesse Kallman breaks down how Danti AI fuses satellite imagery, open-source intelligence, maritime signals, and sensor data into cited answers in seconds.
Maggie 00:36
On this episode of the Mission Matters podcast, we sit down with Jesse Kallman, the co-founder and CEO of Danti AI. Danti is building the future of AI-enabled intelligence analysis. Its platform runs a team of AI intelligence analysts to answer questions by integrating and parsing through diverse data sources, including open source intelligence like social media data, geospatial intelligence like satellite and drone imagery data, sensor data, and more. Their product is used by customers across the US government, including the Space Force and National Geospatial Agency. Before founding Danti, Jesse had a long career in the defense industry across both startups and defense primes, including Northrop Grumman, Booz Allen Hamilton, Airbus, Xwing, and more. Jesse, thank you so much for joining us on the Mission Matters podcast.
Jesse 01:27
Thanks for having me.
Maggie 01:28
So first question, a little bit of a softball. You all of course have .ai in your URL, which is very hot these days with startups. Everywhere you look, people are talking about AI, AI agents, both inside and outside of the US government. So I wanna ask, what does AI agents for defense mean to you, and in the actual context of Danti, what all are you actually doing with AI?
Jesse 01:56
Yeah, it's a great question. We get asked about this so often. Sometimes people even laugh when they hear the phrase AI agent because it can mean so many different things to so many different people. And so from our perspective and how I think about AI agents, it's essentially tools that can execute tasks that are sometimes quite complex, in a sequence. And so they understand how to run the sequence. They understand the steps of the sequence. They understand, based on key phrases from a person, what they're trying to accomplish — how to actually go and do that work. And so how Danti thinks about that in the context of intelligence analysis, or just supporting war fighters overseas that just need key information, we like to bring a variety of different tradecraft and skill sets into these AI agents that understand, based on a person's question: is this a maritime question? Are they trying to solve for a natural disaster? Is this a border issue? Or is this just a simple data request that can be solved from a summarization of 50 tweets, two news articles, and a coincidental image? And so that's how we think about agents — tools that can go, once tasked, and execute multimodal or multidisciplinary tasks for the user.
Maggie 03:10
So perhaps by design, I think your average person is not super familiar with what an intelligence analyst in the intelligence community or in the US military actually does. So can you paint a picture? What does an intelligence analyst job look like pre-Danti, and then where does Danti fit into their workflow?
Jesse 03:30
Sure. So a lot of our team comes from the defense and intelligence community, and they've worked in many different agencies across many different disciplines. I myself have seen a variety of different disciplines, but I think it boils down to a few key things. It comes from: someone somewhere needs information for some specific purpose, and something like a request for information, or an RFI, will come in to an organization — says, "I'm looking for X," or you're providing a briefing to Y. And typically that analyst will, one, take that request, and based on their prior background and knowledge and their level of experience, will know: to solve this and answer this question, I have to execute these steps. I'll typically go across a variety of different systems. So let's say in the case of a GEOINT analyst — they're being asked about activities in a port overseas. What they'll typically do is, that port gives them a key buzzword, where they'll know: okay, we have things like port boundary information. What is the actual boundary of that port? What's encapsulated there? What other data sets? Do we have any visual imagery data sets? Are we talking unclassified, classified? Say, at the unclassified level, what data repositories do I have access to? Do I have the logins that I know, based on my background, I can now go and pull information? The next thing is more of that temporal context. When are we talking about? Are we talking yesterday, two months ago, three months ago? Is this a pattern thing, or a series over time? And they have to start to build that temporal context as well. So now they've got the spatial, they've got the temporal, and then they start sourcing. So let me pull together the key information that I have across that spatial-temporal context, and this can take days in many cases. And once they've pulled all that together, then they can begin the analysis work. So what's actually happening here? What am I observing? What's happening in the surrounding areas? And can I determine what it is this person is ultimately asking for? And if I can, I then bring that all together with supporting evidence into a report that's then delivered. And so that's one example of how we've observed it over the last many years.
David 05:38
Awesome. Well, maybe taking it from the abstract level — SpaceWorks just named you a success story, I think last month. So maybe go into a little bit of detail about what you've been doing with Space Force and various combatant commands, and how they're leveraging Danti today.
Jesse 05:55
Yeah, great. We were happy to see that story come out on some of the great work we've been doing, initially across the Space Force's TAC-SRT program. It's since been renamed to, I believe, BATS. But—
David 06:08
And what does TAC-SRT stand for?
Jesse 06:10
It's Tactical Surveillance Reconnaissance, I believe. I forget the exact terminology. But yeah, it's supposedly for responding quickly to combat and command needs for commercial data that's publicly available and publicly shareable with allies and partners. And some of the key use cases there are things like supporting global natural disasters, maritime activities, border conflict issues. But anywhere the US really needs to be able to provide this information at a fully unclassified level to its allies and partners in the region, so that the US can provide that information to support whatever needs there might be. And so that's been the primary focus of that program. I think it's been pretty revolutionary, being able to bring in so many different commercial vendors — whether they be imagery providers, analytics providers, analysis services companies — to, again, take in-turn requests that could take a week before, to maybe a couple days. And where Danti's been supported is: now, how do we take that request and potentially automate it all the way, initially? Or how do you understand the data landscape before you go and execute those tasks? And maybe to give you a specific example, since this is recent, with US SOUTHCOM. So the earthquakes in Venezuela — that was something that Space Force supported pretty heavily. One, in providing a lot of information very quickly to local NGOs and other governments in the region. Obviously not able to provide classified information, but a lot of their requests are: what are the status of the ports? Are there slips to bring aid in via ship? What's the status of the airport? Are the runways destroyed? What sort of hospital infrastructure — what does that look like? And so what Danti provided in that case is, an analyst at US SOUTHCOM, TAC-SRT, the Joint Commercial Operations Cell, can just come in, in natural language, type: "What are the status of the top three ports in Venezuela in the last forty-eight hours?" And the system, in a matter of seconds, will synthesize thousands of news articles, tweets, information across the Internet. It'll cross-correlate what it finds when it determines where those ports are. I mentioned before pulling that information on the port boundaries, so understanding where the ports are. It'll look at broadcasting ships via AIS. It'll pull coincident imagery, so you get a visual understanding. And then it provides a simple synthesized summary that someone, in a couple clicks, can put into a briefing or PowerPoint slides and send as an email to an NGO. And so that's what Danti provides — that ability to really automate across those different government and commercial systems and get the user directly to that end report they need.
David 08:51
Maybe just tell me — but for Danti, in the case of this Venezuelan earthquake and being able to pull in a lot of information quickly, how would this otherwise go down? What are the analysts using today without Danti, or yesterday without Danti?
Jesse 09:11
Yeah, I was actually there in the ops room at SOUTHCOM for some of this as well, and it was interesting to see the ops center, where you have a bunch of different teams — mostly staff from different parts of SOUTHCOM, from mission partners, other agencies — and they were fielding requests either by phone call, WhatsApp message, whatever the form, and they're bringing that into the ops center. The ops center is typically, again, very manually going by and saying, "Okay, based on this request, who has this information? Who can get on the phone with another partner agency and bring it together?" And it was being pieced together by different commercial companies, contractors. They have a common operating picture up in the ops floor where they're integrating different datasets to see if they could do it there. And so that's how it's been done. Where Danti provided value was, we showed up with a lot of automated integrations to a lot of the systems they were pulling from already, in a lot of the modalities: air traffic data, maritime data, imagery, open-source intelligence. We would simultaneously take those RFIs coming in, also type them in to see if the system could actually just spit the answer out completely automated — completely alleviating all those people having to waste time calling across their different networks to piece the information together.
David 10:29
So, Jesse, maybe building off of the work that you've been doing with SpaceWorks, talk to us a little bit about some of the international cooperation and working with some of the combat commanders. It seems as though international partners are starting to leverage a lot more of the unclassified data that's coming off of a lot of different capabilities, whether it's commercial GEOINT providers or otherwise. And just how do you see the international market shaping up?
Jesse 10:58
Yeah, I think that's a great thing, and it's probably one of the biggest growing areas we see in general right now. And maybe a more US-centric example would be — we actually did a very rapid deployment with USSOCOM, with NATO, during an exercise called Trojan Footprint this past summer, where we got contacted one morning and there was a need for equipping a lot of allied forces in Europe with information tools, but there was nothing that they could do quickly. And within two or three hours, we had a deployment available for SOCOM that allowed them, plus a number of different NATO countries — I believe it was about eight or nine — and all of their special operations forces to connect unclassified information from open source, new social, imagery, a variety of different things. And now any SOF operator in any language could come into one terminal and ask a question and get provided the entire coalition's information in seconds. And we saw anyone from Polish and Hungarian to French forces just get dropped into our software and start running things like GEOINT, SIGINT, OSINT workflows all together. And I think, more broadly in the international market, this is a function of — most countries do not have the infrastructure the US has. They don't have tens of thousands of highly skilled analysts sitting in offices waiting to support and answer questions and fuse information. A lot of these foreign partners know the information exists. They want control. And I think what we're hearing from a lot of the space companies — a lot of even companies like Palantir have coined this data sovereignty thing around the world — is that all of these international organizations now, they want their own data capabilities. They want their own satellites. They want their own feeds to the key information they need, but they just don't have the massive trained analyst force to do it. And I think that's a great opportunity for companies like Danti, where we're already providing a lot of that encoding of when is it the right time to use a synthetic aperture radar satellite versus a bunch of tweets, versus the beacon off a ship. And so how does someone like the Korean government or the Philippines government better leverage all of this information with an analyst — doesn't really matter how well trained, or their background, whatsoever. They're able to control their data environment and control their own information that's getting out to key users.
David 13:20
That's awesome. I guess a lot of what you're talking about, with all this data aggregation and fusing, I will say, seems a bit reminiscent of Palantir, and I do feel like they're kind of an 800-pound gorilla across the software landscape within the department. How do you view Palantir, and where does Danti fit into the tech stack or ecosystem of tools available to end users?
Jesse 13:47
Yeah, I think Palantir obviously has been around a long time, and I think they've done a lot of amazing work in revolutionizing how complex organizations connect silos of data and build useful applications on top of that, and coining some of these phrases like data ontologies and how do you build relationships. So I think they've really advanced how a lot of different applications and different folks in the community can operate. And I think where Danti fits in — we, as a company, provide capabilities, one, that help organizations do a lot of these data integrations connected in more of an automated way, where you can just tell the system what this data is. We provide a lot more in-depth, very detailed workflows for that end user, who is likely not going to use a more complex system like a Maven Smart Systems or something like that. But really, we deploy within that ecosystem. So today, we're actually even co-deployed with Palantir in organizations like the US Army, and in that Venezuela example. We are able to feed information into those common operating pictures like a Maven Smart Systems, where those are really powerful views of the entire visual data landscape. Whereas in Danti, you're asking an agent — essentially an analyst — to start running cross-modal workflows for you, doing a lot of fusion, doing some of that analysis, to then spit it back into those common operating pictures. So I think Palantir has laid the landscape to making a lot of those data feeds more easily integratable and common across the entire enterprise. And I think it's paved the way for a lot of companies at the application layer like ourselves, who are doing a lot of great work, to take the last step with a lot of these organizations.
Maggie 15:40
So Jesse, from the use cases you've been describing, it sounds like you're able to work with a number of different combatant commands, a number of different services on different use cases, but you're able to use one platform for all of those. So can you talk a little bit about how Danti is flexible and able to work on each of these different tasks?
Jesse 16:01
Yeah. I think we talked a little bit about some of the bigger players in the space, but there's a lot of organizations that don't need that level of infrastructure rebuild, where, again, we can step into some of these mission partners who have a very simple... I mentioned maritime domain awareness, where they already have some of their own infrastructure, their own ecosystem of vendors. Danti will allow them, in a matter of hours or even a couple days, to completely connect that disparate information data environment, and then have someone of any skill level now be able to run the workflows of what were highly expert, 15-year subject matter experts. And now it alleviates those subject matter experts to focus on those really difficult tasks, like the ones that absolutely need a human spending hours researching and analyzing. We can now alleviate that, so we provide that data integration infrastructure layer for them. We provide all the workflows on top of it. We provide a lot of the encoded tradecraft so that, again, anyone from a mission partner that can't spell, or doesn't know what SIGINT means or OSINT or any of those things are — they're able to run those workflows across those different tradecraft areas.
Maggie 17:13
And then expanding that further, so CENTCOM can use this, PACOM can use this, but you also have commercial customers using this. Can you tell us about some of the commercial use cases for a tool like this?
Jesse 17:24
Yeah, just like the US government has issues connecting highly disparate and very spatial, anchored-in-the-real-world datasets, so do insurance companies, so do commercial real estate companies, so do global logistics companies. And so some of the things we've seen are anything from, again, global logistics companies that wanna understand impacts to their facilities due to ongoing conflicts around the world — where cargo may be stuck in the seas due to conflicts outside of Iran — to even real estate investors that are trying to understand the complexities and nuances between things like foot traffic or traffic data around a location and how that relates to demographics of a local neighborhood or population changes, or even the structural information of a building, and how do you use all of that information together to help with things like real estate acquisition that today also require a team of expert brokers to pore across this information. So that's how we think about it, where you have this anchored-in-the-real-world problem that is a huge mix of information across different disciplines, where you're just trying to help someone use all of it interchangeably to answer whatever their key concern is.
David 18:42
So maybe, it seems as though the use cases for the military are not so dissimilar from those on the commercial side. What does the setting-up process look like for a commercial vendor versus the department? Maybe we'll circle back to the department, and you can provide the listeners some of your trials and tribulations and what it's been like. But just curious — how have you seen the differentiation there?
Jesse 19:11
Yeah, I would say, for the differentiation on the government side versus commercial — in the government landscape, there's a lot of security restrictions, data access restrictions; there's a lot of sensitivities into where that data can be shared. I would say it's not extremely dissimilar to commercial, where a lot of enterprises are also thinking about how do we secure our data. There's a big topic now around data sovereignty for both governments and enterprises. And so in the enterprise market, a typical startup, or a typical new start — a lot of times these organizations are just using our software out of the box. It's cloud deployed. It's got tons and tons of information around the world. They're able to get comfortable with what the system can do before they start integrating it with their own infrastructure. So, being able to monitor global facilities to start, you just need to tell the system where the facilities are. We don't need deep integrations with their own enterprise systems, but we've done other applications where insurance companies can directly either integrate or run automated exports of things like policy data, more sensitive commercial or customer information, into the system, because we deploy the system in nodes where every organization and every customer's data stays secure locally in one environment. They have complete access and control on it. It doesn't send data out to external AI models or anything like that. Everything is locally deployed and secured, and so they can be assured. But anyway, they can start with a more public and open system, but they can also have even their own on-prem deployed systems behind the firewall.
David 20:54
Very cool. And so you've obviously had a decent amount of success with the military contracts. What does that look like on the networking side, and getting access to government network information and data?
Jesse 21:08
Yeah, for those that listen to this that are maybe also looking at getting into deploying software at the government, it's not an easy road. There are so many things required — from facilities clearances, authorities to operate, being able to actually have the credentials and do the deployments. We joked earlier today about even having CAC cards sometimes to be able to access certain network terminals, and sometimes the hurdle's there. And so we've been around a little over three years, and we've been able to get through a lot of those hurdles, where great partners like the National Geospatial Intelligence Agency supported us to get our facilities clearance. So now the company itself can hold classified contracts. We can have cleared staff that can be cleared to go into buildings and access the right networks and systems. The system's already gone through provisional ATOs, and that's where you essentially have the government review the security infrastructure around the software and how it's tested and its deployment, to even something that sounds as simple as getting military IDs. But a lot of these terminals require all of these things to be able to then just deploy your software directly into that government network, to inherit the controls of that network, to inherit things like user rights and permissioning. And that's critical for an application like ours — to know when someone puts their ID card into a network terminal, the system is gonna provide them only the data that they have access to see, and they'll only be able to leverage it in a way that keeps that information secure. So we understand why the hoops are there, but yeah, for a lot of new entrants, it's quite difficult to navigate.
Maggie 22:46
Yeah. What's remarkable to me — I saw genai.mil recently got stood up, but if you actually look at the models on the back end of something like that, I believe they're models that are two or three years old. So the department is just now getting access to the ChatGPT that we all had access to three years ago. Which does bring me to another question, which is: how does something like Danti compare to a ChatGPT or Claude just straight out of the box? How much are they able to do versus what is the special sauce that you all are bringing to the table?
Jesse 23:16
Yeah, it's interesting, 'cause when we deploy in different environments — we're just deploying in a different environment now with the Army, where I think in that case it's an OpenAI model, where in some other cases it's a very different model. Our system is architected to be able to leverage the best as these models improve. They get better at information synthesis tasks, or they get better at whatever the task might be. We can leverage that model, but there's a lot of parts of our system that don't require LLMs at all. Some of it is graph technology, some of it is learning graphs that are really built on Danti's own IP that can understand and learn this tradecraft. Whereas the models that are just so broad and trained on a very broad set of data — once you start to go into these physical-world environments that have much more nuanced and complex data sets, things like signals from satellites and organizations like a HawkEye 360 — this is complicated information that is super powerful. A lot of these models just have no prior training or understanding of how that data is used and how it's used across a variety of different tasks. And so what Danti does is more of the orchestration — based on what the user's trying to accomplish, understanding what the tradecraft and use is there, and knowing when the right steps to introduce a model are. The other thing is, there's very little tolerance for hallucinations or incorrect information. And so we have to provide a lot of standardization in how results are presented back to users, and citations. And so you can only introduce some of those non-deterministic models at certain parts in those sequences, and knowing also when you can't use those models is also very important. So we provide a lot of that, so that the end user shouldn't have to care nor think about that. They can trust that when they go into a system like Danti, they're asking their question and they're getting a cited-back answer every time, consistently, knowing the provenance of all the data — data that is their government's information living on their government server completely securely — and then just leveraging the AI to speed up certain portions of that task.
Maggie 25:21
So Jesse, looking back at your career, you've worked at defense primes, you've worked at startups. From all this experience, what do you think startups are uniquely good at in the government space versus where do you think more traditional primes will still win? And maybe a secondary question: how do you think about opportunities for startups to partner with primes to accelerate the chance of bringing technology to the warfighter?
Jesse 25:45
Yeah, I think that's a really important question, myself being a Booz Allen alumni and seeing all the great work they do across so many different agencies and their ability to have that deep customer understanding. They're involved in them and many others. You guys mentioned some of the deployment environments. They're supporting a lot of the build-outs of those environments as close, trusted partners of those organizations. And for a startup like us to just walk into an intelligence agency and pretend to understand the complexities and nuances of their IT environment, their procurement environment, the user infrastructure, and the applications and workstations? These defense primes have already built it. They understand it. They understand where the issues are, and they can help accelerate companies, I think, much more quickly versus a company trying to come in by themselves and do everything new. I think there's definitely some nuances in things like shipbuilding or autonomous aircraft or things like that, where you're building a complete end-to-end system. But sure, they do have C2 environments and others that maybe some of the primes are engaged in. But from a software perspective, it's really, I think, important to leverage a lot of partnerships with those prime contractors, again, because they so deeply understand the customer environment.
Maggie 27:04
Yeah, absolutely. So you've been at this now three years. You've been working in the defense space longer than that. What advice do you have for startups embarking on a journey, the kind of journey that you've been on? Or what advice would you have for yourself if you could go back in time and talk to yourself three years ago?
Jesse 27:21
Yeah. Don't underestimate how long things may take. There's a lot of buzzy news stories out there about companies getting massive contracts and going huge, and I would say that that is not the standard. And I would say, maybe look into how long ago those organizations actually started. And I think my advice to folks is, definitely over-index on the amount of experience you bring into the team early on. Don't think that it's just simple — like a commercial industry — to go and build something and disrupt everybody, and you're gonna grow like a Snapchat would, but inside the government domain. It's definitely a different animal. So heavily rely on a lot of experience as soon as you can get it. Relationships are critically important, so make sure that you over-index on those things as well. And just make sure that you plan for the right time horizon, because things are complicated. Things take time. There's things outside of your customer's control. It's not like in the commercial market, where your customer can just like you and buy it. There's so many other things behind the scenes that they cannot control, and you just have to be patient and understand that there's just frustrating things that you're gonna encounter that are well outside of you and your own customer's control.
Maggie 28:36
Great. Well, Jesse, with that, thank you so much for coming on the Mission Matters podcast, and we're super excited to be investors in Danti.
David 28:42
Yeah, thanks, Jesse.
Jesse 28:43
Thanks for having me. It's great to work with you guys.