Drones And AI: The Future Of Defense | Startup Garage Show #7

hello everybody my name iska and this is star top garage show as you can see we are right now in the in the garage and everything is beginning from the garage so today I'm very excited to announce our first guest cow Orchard and his startup he will tell about he started by himself for now I just can say that I've met Kyle not so long time ago but they already had a very interesting conversation about different things about Hardware about software about Knowledge Management about artificial intelligence and that's how we came to this moment and we finally can discuss it even deeper uh those things it's Kyle is interested in and working on and yeah it's I hope it will be exciting exciting um conversation about drones and how artificial intelligence can revisionize their hands with Knowledge Management so yeah Kyle tell a few words about yourself yeah arson thank you so much for having me and yeah I guess I'm the first in-person guest on the startup garage show so thanks again for making it happen we have excellent equipment and yeah so my name is Kyle I am an early stage startup founder located in San Mateo and what we do is we're building GPS denied drones for creating Bim models out of buildings that already exist or are under construction so essentially drones that can do scam to Bim so and you know it's not an easy undertaking but we're working steadfastly towards that goal and yeah I'm happy here to to talk about defense and how computer vision Knowledge Management Knowledge Management and drones play a part in in you know defense in other areas it's it's great Kyle uh you know before we will go to some deeper level yeah I know that you're a student of still playing I I really respect this this man because uh what he's doing is it's for me is one of the best approaches in entrepreneurship and now you also start your journey as entrepreneur so what can you tell Maybe about your experience on what stage are you right now and what what do you plan to do next somehow our community can you know help you as well yeah yeah so uh to give context why you know why I'm even giving you a two cents on on defense is because my company started out of hacking for defense at Stanford so my last last quarter at Stanford when I was an undergrad I took a class with Stephen Weinstein Steve blank and Joe felter and these are all guys that are aligned with the Hoover institution and the Hoover institution recognizes the need for driving innovation in the private sector for the Dual use case Paradigm essentially to transition hard tech and software that not only can be used in commercial markets but can also be used for our armed forces and you know for our defense so essentially the the Crux of what Steve blank professes in class is you know get outside of the building you are ignorant and pretending to be an expert when you're not is the easiest way to to waste a lot of resources money time so essentially customer interviews and having a rapid iteration on your MVP not spending too much money on your MVP but having maybe a list of features a very low Fidelity prototype and showing that to as many people as possible and figuring out how much they would pay for it is is a huge way of de-risking you know spending a lot of hours and Cycles building something before it's actually validated and the customers actually want it so he taught us this lean launch pad methodology of going out and doing customer interviews in 10 weeks we did over a hundred uh with 25 different organizations and back then we were working with the US Navy and the Office of Naval Research on automating ballast tank inspection and we we definitely we validated the need we talked to the key stakeholders the beneficiaries and we presented a recommendation on how to improve operational Readiness and the availability of Warships for the U.S Navy um netting a 20 20 increase in operational availability of these ships meaning that 20 20 more available at once because of less dry docking time due to our solution so he taught us essentially like you know just to summarize all that you don't know anything unless you get outside the building and just do those customer interviews and don't burn Cycles unless there's a reason and worth of doing so it's it's awesome Kyle I I you know I've read uh several books of Steve blank and I found it very useful and it's very wise decision you know to start with such message and yeah that's great that you have so so so so so so nice teachers so I I'm sure you will you will do a very very great startup so just yeah keep going and yeah it's great so why have you chosen exactly this area that I'm working right now so what what brings you to this moment yeah yeah so to drive to provide a little more context we started off as a computer computer vision company and essentially we were told that Hardware is hard we should stay away and you know when when you're starting off with startup early days you know cash kind of strapped entering into Hardware is a lot of people think it's suicide so then we started off with a computer vision angle you know okay let's abstract away the hardware some other manufacturer that has already you know been producing you know remotely operated Vehicles let's have them worry about that end and then we'll worry about the analysis software from either the video or scanning that these remotely operated Vehicles would do inside of the ballast tanks and then you know we did some more market research after the class and we realized that you know in the Drone space um there are computer vision platforms out there annotation platforms inspection platforms uh B2B Solutions and we realized that this is already kind of saturated um you know there are there's this company called Genex that does crack detection in bridges and Roads and there are various other companies that that provide these software ecosystems for categorically doing a power line inspections cell phone tower inspections doing a photogrammetry orbit creating a 3D model and then being able to pin annotations on any part of that 3D model and we realized you know what we were doing wasn't that different software uh in this space this computer vision slash annotation slash inspection space geospatial awareness that that was already kind of solved what really isn't solved is the GPS denied formula for drones drones have a well-defined behavior in outdoor space because of the satellite constellation being able to provide you know relatively one meter of accuracy for drones when you get into more specialty applications like lidar scanning and surveying using drones that one meter accuracy is is not not good enough you would need something like rtk or PPK to get centimeter centimeter level accuracy for those applications but GPS denied um flight is still an open area of research it requires Cutting Edge sensors perception and slam algorithms simultaneous localization and mapping techniques which is you know heavily used in robotics so pivoted away from computer vision we were first told that Hardware is too hard then realize that the software that we wanted to build to help inspectors of the US Navy already existed and since then we've focused on doing things that are not readily available and for that we've decided on just working on drones that can I can scan buildings really well with a high degree of accuracy such that blueprints can be recreated such that a accurate as is condition of buildings under construction could be done rapidly you know could be done every day could be done every week such that a construction manager doesn't have to walk around we if we looked at that solution and we were like wow that is something that is not readily available all this software we're working with thinking of making it was already out there so that's that's why we're we're you know why we're doing this um and we think that has a higher risk reward trade-off but we think ultimately the effort will be worth it yeah it's exciting you know I I I I was I had some experience in this area as well from my very first education and I should say that this kind of problem is very very uh important in even in industrial sector as well so where you can you know make a very precise models or not on the buildings but also some industrial objects there are different ones so yeah I I see a big opportunity for this technology as well and yeah that's awesome that's you already started and yeah I think you you will be successful yeah hopefully so okay thank you for such great story and yeah now maybe we will just go to some other topics it's close to it maybe not too close but somehow related as now it is you know I'm from Ukraine and we have this whole scale War right now and drones are taken it's so important dramatic role in this world Modern Warfare so it's used for for for different you know purposes you know for for prospecting for for yeah investigation for for uh even for for uh uh war war a talks I don't know any kind so yeah drones became very important uh in every in every area and of course it's becoming more advanced because I remember when first we were doing our craft drones brother copters and other types in hunger space it was very amateur but it was really exciting and now there are so many Advanced drones and produced by different brands and yeah it seems like this technology is evolving every every year so let's try to understand uh yeah the right yeah the rise of hard tech startup as you said it is really the most and the hardest um the hardest area I mean it's much easier to to create some software startup and Hardware because Hardware yeah probably you will tell us so let's try you know to figure out where it goes and where where this movement what is the direction of this movement yeah so with with hard tech the United States is is notoriously behind um when it comes to drones so DJI was started I think roughly a decade and a half ago two decades ago and it was started in a dorm room no yeah and um yeah I mean I think the the cost of Labor and the cost of manufacturing in the states had a big part of discouraging hard tech Innovation and there's been a lot of Outsourcing of hard tech manufacturing and you know you see it with a lot of products here in this country have you know design in California manufactured in China and I think there is some sentiment of bringing it all back the manufacturing jobs but arguably I think um that's a that's a big undertaking to bring everything back because the system really relies on on a lot of Outsourcing so I think for a little while longer we're still going to have the design of California made in China kind of you know philosophy but it's starting to change which is which is good because you know in the case of of the pandemic we saw how the personal protective equipment the PPE was was kind of they could hold that hostage against us it's all made in China and then every country essentially during a pandemic enters an isolationist kind of policy right so if we rely on on on foreign Hardware that's critical to National Security whether that's chips AI let's say gpus if they're manufactured in in China I know that Taiwan the Taiwan superconduct TS what was it Taiwanese superconductor company um they um they they are they have a huge market share for for the pcbs and all the critical components that that are put into cars and then computers and all sorts of different kinds of electronics so having that all in other countries such as China and Taiwan Taiwan is on the brink of invasion and China at any point can can just dictate isolation as policies we're at huge risk and we need to be very cognizant of that so having some kind of diversification and and how we Source our critical equipment that would go a long way in preserving our interests especially um NATO is starting to heavily rely on on drone combat and circumventing that kind of you know the Iranian drones and and different kinds of evolving rapidly evolving types of attacks that we've been seeing on the battlefield so at any point you know with with with Russia and China starting to align um yeah I mean bringing back the manufacturing here would be critical you know I'm I'm totally agree with you and uh I should say that uh some of my friends they uh also from Ukraine and they started the converse startup like 10 years ago and uh yeah the story of of their startup is it also was Hardware so a very very typical story so they had a sales office here in the United States uh Manufacturing in China and but designed in in Ukraine so it's a bit different but very very similar uh so yeah it seems like China is still very important for the hardware production and yeah it's it's I think it's natural as you know all this process was was made not in one day and it was made for years so uh but I believe you know if you will look at what Elon Musk is doing I suppose it's the best way is not just to bring a manufacturing bed but all but also uh improve it and create it so like more usage of Robotics more usage of not human labor but robotic labor because it's more technological it's more sustainable so that's how you can produce a really high quality products and uh in in Moss production as well maybe sometimes you don't need some massive production maybe you just have to to create some something in a note not not a so big scale because especially if it will be prototyping properly robotics is even better because if it will be used for uh like unique uh unique uh prototypes maybe it's even better I I don't know it's just an idea so but that's yeah I'll return back to the to the drones and um you know now computer vision is very important and it's it's using for analyzing the the object and how do you think um um and yeah that's how it's connected to artificial intelligence to Knowledge Management how how do you think this computer vision is reshaping differences strategies and offering some unique Solutions so let's discuss it yeah yeah so computer vision is is playing a very big role in Automation in in processing area footage aerial imagery so with the with the proliferation of DJI drones in in the conflict in in Ukraine there there's a lot of manual operation of these drones so that means an operator has to be you know within several kilometers of what whatever area of Interest um is being either surveyed or or either targeted so and you know there's there's inherent risk with with operators being so close to um the areas of Interest especially in that conflict um and then also um in cases of where GPS is is compromised or or jammed GPS deny navigation algorithms can play a huge role in helping helping these small unmanned aerial aircraft land and take off from the right places and and you know if suddenly there is a jam while an aircraft is is in Flight Can it can it use the the perception of its surroundings to still get from point A to point B essentially so computer vision has made great strides with these kinds of tasks and also um with Target acquisition also with um and you know non-militation non-militarily simple tasks such as Counting Cars in a parking lot Counting you know bales of hay on a farm uh measuring uh crop yield measuring dead areas where um you know maybe you added a little too much there's too many nitrates in the soil too much phosphorus um you can better allocate your fertilizers better allocate your water and you know satellite imagery has multi-spectral capabilities um but you know you have to satellite imagery is the cost of acquiring satellite imagery is decreasing but it's it's not readily available for everyone you'd have to be like a small to medium or I'd say medium to large Enterprise to say you know knock on maxar's door and say Hey I want to have satellite imagery from yesterday that's not available for everyone um you know if you're a government agency much easier but let's say you're small business you're a farmer your who knows like a property manager it's much harder to acquire satellite footage from yesterday or from even a month ago and you know Google Maps and Google Earth and all the mapping platforms they update roughly every couple months and and they don't equally update every sector of the earth they update areas of high population first and then and then they eventually get to the rural areas and then all that so computer vision and drones gives a unique opportunity for for small smaller stakeholders to take control of aerial intelligence right and not having to manually annotate and manually write down things that they see in their footage um you know there are there are there are companies now that can automate these tasks that I was talking about these small practical tasks that are done with the pair of eyes in the sky and then maybe attached to an edge device an edge device that can actually process that footage and and maybe equipped with a GPU that can they can run object detection segmentation localization and you know techniques such as that um you know these smaller smaller businesses can can actually have very powerful tools at their disposal and a lot of these things are open source you know you can put you can put YOLO the you only you only look once algorithm on on an edge device and you can count how many cars are in your parking lot and you can see how to better design the next lot if you have another lot that you want to build or you can better analyze uh what kind of you know what kind of customers like you know kind of vehicles are entering your lot and then from there you know like you can estimate other things um yeah but but yeah I think it's exploding in the amount of possibilities and it's a very exciting time yeah you just remind me a very funny story you guys just told me uh yesterday they were in some bar or disco and uh they said we never seen such a lot of people that's so weird but in a different way so I mean that's sometimes computer vision can be used also for analytics who is entering the you know your bar for example who is your target audience yeah it's also it's also uh yeah reminds me this this funny sensation so let's go to the next topic drones in defense from surveillance to combat yeah it's a very very hard topic but I think it's nice to discuss at least what's happening and uh let's just try to feature startups that's leveraging drones equipped with artificial intelligence and computer vision for military applications yeah you already mentioned the GI probably they didn't want to to make it used for combat but it's what's happening but yeah let's discuss it if you if you if you want yeah so DJI issued a statement that they didn't intend for their their drones to be used in Warfare but unfortunately that's just what happens technology is is itself amoral and it's it's used by by people and and people that can have nefarious intentions and you know that's just the State of Affairs currently and yeah I mean right now um there is a significant change in conventional Warfare essentially so if you look back all the way to the Revolutionary War in this country we had our revolutionaries the Continentals they changed their style of warfare against the British you know the British they were a very and they they essentially marched in matrices right yeah you know with let's say you know a few rows and and many columns and the rows would you know the first row would would duck down for the second row to Fire and then the first row will come back up while the second row would fire or or would reload rather and you have this very orderly system of combat um Harvard when when you factor in guerrilla warfare tactics that some of the Continentals learned from the Native Americans the British really really struggled in certain scenarios because the way that the British fought other Empires was a very it was kind of like a very gentle gentleman-like uh affair right you know you have two armies and they would just be these two like two-dimensional grids that would that would fight each other and then um when the British faced some of the Americans once they started getting smarter um they were really suffering heavy casualties especially when going into forests and and on on convoys and stuff like that so anyway um changing changing the the way that you fight um when when the circumstances are against you that's that's not that's not something new this has been seen throughout history and drones are are a new form of guerrilla warfare essentially yeah that's true um uh yeah by the way ukrainians also uh I mean kozaks in the past are also very inventive in in the world without Warcraft and uh yeah as you as you mentioned yeah many of European armies were very straightforward this was the time when it was sufficient probably because it it were more organized you know so if if you were attacking Librarians and and they're very organized military forest and of course organized will be better than their you know chaotic one but then it's it's changed a bit and yeah so I'm agrees it's uh it's completely changing all the conventional Warcraft so um let's try to Showcase to Showcase how these drones are enhancing surveillance capabilities and enabling precise Target identification and stream streamlining data collection in defense operation I know for sure it's now if Ukrainian Army is using a very efficient systems sometimes it's like craft one and maybe not some big scale systems but this is what makes Ukrainian Army more efficient so yeah let's try to Showcase all these drones enhancing surveillance yeah yeah so essentially you know you can you can be in a very secure location you don't necessarily need line of sight of the Drone and then you can just send it over like an entire entire forest or an entire field or across various you know difficult um difficult terrain essentially and then you can I guess you guys probably already know how this works you just fly the Drone around uh you send it back you take notes on what you saw and you know and this this kind of Knowledge Management is is not the most efficient but it's it's effective right for for very localized operations you know either something is there or something's not right uh but for more complicated tasks such as analyzing movement of troops uh you know various different movements and and and Paths of convoys and Paths of military equipment and you know if you have a lot more that you're tracking than this kind of pen and paper annotation doesn't really work as well and and DJI does not design their their software for these kinds of use cases um if if you take a look at what the use cases are for the DJI software it's it's you know simplified um you know movement that's you can do um automatic orbits you can do um you can you can do a path planning to a certain extent for mapping purposes you know for the Enterprise drones but you know for you know entry-level models like the mavic or or other entry level models the point is not reconnaissance yeah the point is to record 4K video to show to your friends to record shoot a family video shoot a music video you know maybe do a tourist video for your for your hometown but these drones the intended use case is being completely flipped on its head and that's where Knowledge Management and geospatial intelligence could could really come in and play however DGI is not going to create software for this purpose because they uh they denounce the use of its use for violence so this would be something that would need to be built on top of their sdks or their their apis and I'm not sure um whether the ukrainians are doing that but they're very you know very intelligent hackers essentially I'm sure they have yeah for now it's just a matter of of survival so yeah that's why you know some unfortunately some of my friends now have to to point so they have to be very inventive because it's about their life and this is the point yeah that's I want to discuss next so if the potential ethical consideration and challenges associated with as I say that with the deployment of autonomous drones incumbent scenarios so for now these drones are controlled by people and of course if artificial intelligence will be in controls it's where it is ethical consideration and challenges appear so yeah it's it's a bit scary because you know it's it's almost like uh we were watching in this Terminator movie even drones and Terminators were were fighting against humans and yeah it seems that we are very close to it and drones became the first real to ol that's used for this purpose yeah yeah yeah so one thing that computer vision will struggle with is when when you have to identify targets how do you know if it's Friendly Fire how do you disambiguate between a hostel a friendly or a bystander yeah essentially so the problem is with with the development of these kinds of algorithms and and especially if you're under desperate circumstances it's very difficult to to ensure the the ethics of AI systems in such a necessary complicated Battlefield however you know in in these cases you know where where it's total war essentially you know civilian areas are not untouched right it's everything's kind of Blended together and that that's one of the terrible things about the war um so yeah like the use of AI and and this kind of this kind of theater this is technology that should take years to develop you know in terms of if we're talking about rolling out Target automatic Target recognition on on DJ on consumer grade drones yeah right um but right now that's not really happening what's happening the common attacks are are just attaching grenades or attaching explosives and then um going to tanks going to like high traffic areas and and hoping to you know eliminate the targets right and hoping the tank hatches open and flying directly in and these are Kamikaze style attacks but there's to my knowledge I haven't really seen um a high level automation with regards to those kinds of attacks I've only really seen operators you know hidden by an Alcove or a safe safe area and piling the Drone and you know manually finding the targets and acquiring it but you know my only my biggest concern is with with you know kind of makeshift computer vision algorithm being rolled out on a battlefield on on drones not necessarily designed for for combat you know it could kill anyone yeah someone on your team someone on a bystander and disambiguating that you it is just very difficult yeah yeah we still have to be very Consciousness about how we use it differently yeah it's you know it's scaring me the most because when something is just flying around you and You Can't Hide You can run yeah it's it's so so so yeah dangerous so yeah but as we know not the the it's just a tool somebody have to you know to push the button to pull the trigger yeah that's why it's mostly telling us about the culture about the those who are using these tools so yeah I believe it's till now all the moral and ethics fundamental fundaments is very important yeah it's very important and also privacy as well you know when you have drones hovering around and analyzing your movements like for you know like if we go back to the example earlier that I mentioned of you know seeing how many cars are parked in the parking lot and and then maybe tracking um how you enter the store or how you leave and how much stuff do you have when you leave you know versus how much you came with in with do you have a child um you know do you have a stroller and then once you have these like kind of drones hovering around you know there's already this kind of computer vision uh layer on top of the CCTV Stacks that you see in businesses right you know there is a company called dragon fruit AI that they would they would like kind of monitor um computer that they would monitor uh like CCTV footage and help business owners skip to areas where there could have been a crime committed and amongst other use cases and uh Cloud cloud-based video analytics right so there's already kind of automation being done on CCTV footage analyzing what people do and and um you know how they react to certain products and you know what like what part of the store were they most excited in you know and drones are essentially an extension of of like closed circuit television and certain certain use cases right it's just that they're not permanent infrastructure installed um so you know but the thing is about drones uh they um they can be flown anywhere they can be flown over backyards it can be flown you know very by close by windows and stuff like that so you know um there there are a lot of privacy concerns yeah yeah that's true yeah I like the idea of using uh computer vision for analytics it's what can bring the better to businesses but yeah the Privacy she's still very important so they so it's nice to be sure that this data is not used against people yeah yeah if it's real used to bring more value to the people a bit ethical if it bring to you know blackmail people later that's not good okay thank you Kyle it was a very interesting um Journey so let's go to the next topic it's a high-powered Knowledge Management in defense yeah by the way uh not so long time ago I I was on some um HD HD uh presentation uh in Stanford University and there was some yeah the guy was presenting his work in using AI in military but in strategic level and yeah it was very interesting for me as you know there was a lot of discussion how these systems artificial intelligence systems could be used in different integration different ways but let's try to highlight the importance of the Knowledge Management in defense separation operations and the role of AI in transforming this domain yeah I see a lot a lot of of dramatic shifts what what what can you tell us yeah I recently read about scale AI released a large language model tuned for a defense strategy and defense decision making and I think I forget who exactly the the customer was within the government but essentially the military to this day has a lot of a lot of archaic processes and also still use paperwork in a lot of cases they have come a long way to digitize paperworks reports and stuff like that but what scale AI was doing was essentially ingesting all of all of the reports you know for example like let's say you know officers they have to manage you know a lot of soldiers below them and there's a whole hierarchy of of reporting essentially you know there's always a there's always someone to report to and managing people that entails a lot of paperwork so you know what scale AI is doing is ingesting a lot of the the documents in in some of these um branches of the military and offices and saying hey you know I can provide a concrete clear picture of you know your personnel uh what are Trends uh in in terms of you know like our our soldiers failing to report for Duty on time are are there systemic issues in in terms of the operation of of a particular company or Battalion or or whatever whatever unit you want to call it and saying hey like you know we're going to package we're going to scrape all this data and then we're going to package it in in um large language model such that you can ask it questions you know say hey like how's how's company a doing versus Company B today um okay the the large language model will say okay I'm looking at the attendance or the performance reviews or the reports you know company a is doing Comfort better than Company B for these reasons and having a an interaction like this can actually go a long way for for decision making you know because officers before they would have like file cabinets full of records and the issue is you know it's just a trust you know a lot of these things are classified and and not for um public dissemination and the way that open AI improves its algorithms is is by ingesting all of this and making the model better right it's additional training data because they study the interactions between the user and and the output of of the chatbot essentially right you know does the user react favorably or disfavorably you know was that a good response judging from how the user proceeded that was a good response and Samsung Engineers actually made a critical mistake by uploading some of their source code to open AIS chatbot and and then and then I think other engineers and other companies or or um I forget how I was discovered or we're getting example code generated that looked like it was from a different code base a proprietary code base so that's why um you know the scale AIS I think it was called Donovan or I forget what it was called exactly but what they're doing you know it's very serious in the in the sense that their data ingestion and training process needs to be very secure and it cannot benefit from the same open source Network effect that that the open AI is chatbot is getting you know Crossing they were I think they were the the fastest app to reach a million users in history um yeah and you know openai can benefit from this scale AI in the case of you know providing a military llm cannot benefit from from that as much um so that's just one example there yeah I I really really find it yeah it's probably not funny for Samsung to kill this situation but yeah in general yeah I see it's um AI as a tool is a really great great technology but as I said before in previous episodes it's very important um what data do you use for training and how do you train it and Googles and use this llm models uh yeah so I mean it's it's definitely will go in the direction when in every area we will have a very specific AI that will be trained specifically for this purpose because I hope this will bring us this privacy and security so it seems like yeah of course there will be some technology for training and then for for using this this specific AI for this specific uh problem you know so yeah and of course it's not good idea to mix all this uh things together because it will be it could be very very unsecure okay so let's showcase some startups it's on developing and I've powered platforms to capture analyze and this evening critical knowledge using defense organization I know that somehow volunteers but yeah we are not advertising anybody sorry nobody is sponsoring us yet um yeah probably um yeah I know that there are some Advanced Solutions it's already used and uh uh yeah some of them probably we don't even should know yeah but uh uh maybe maybe some of them so what what what what can you tell us yeah about this I'm not very deep in this area I just heard about volunteer as uh I read the book uh uh from zero to one it's a great book I should say yeah and volunteer was maybe at first used for for different purpose but now it's also used for defense as well um yeah so yeah yeah so I think a couple companies to know are you know obviously palantir C3 AI also Andrew Industries Andrew L has been making a lot of Investments and in startups as well as inkytel these are kind of the big players in the defense space that are on The Cutting Edge side versus the the primes the primes like uh Lockheed Northrop yeah they're more focused on on delivering on those big contracts for you know the tanks the fighter jets maybe spacecraft satellites stuff like that and Boeing as well and then you have those companies that I mentioned earlier you know working on more cutting-edge stuff not cutting edge but you know maybe more software oriented or if you're Vision oriented and you know there there are others um but I don't have huge market research of that area and shield AI that they're developing an AI pilot essentially yeah so a fighter jet pilot not not like you know those those uavs that that look like the the silly um I don't even know how you describe the you know those drones that were flying over the Middle East and kind of um yeah I forget what I forget what it's called Maybe anyway um but developing a fighter jet AI significantly different to program a dog fight versus a point A to point B and and drop drop a payload and point C right um tactical maneuvering and and firing and you know Target acquisition basic Maneuvers all of that being factored into an AI pilot that is that is a significant uh it's much more of a challenge in my opinion uh versus the current you know uavs that that are being used by the military because you have you have people still operate those um just remotely in some small forget somewhere in Nevada they have they have like a little little place where they um pilot some of these drones um but yeah that's kind of a little sample of the landscape right now and everybody's different a good place for such for such works yeah as yeah there are loads of base uh you know I I think it's Ukraine is also will be a very interesting uh player in this kind of technologists is now I know a lot of people start working more than that not so a long time ago there are a new cluster was created like meal Tech so I I suppose a lot of interesting technologies will be implemented in the results I'm I don't know how big do it became but but it's definitely no different year of you know of all events and and basic knowledges could be you know used in real time and all this experience could be could be [Music] implemented in a very short term so um yeah let's come to to yeah to do the latest sub topic so let's explore the benefits of intelligent Knowledge Management System for enhanced decision making faster response times and improved operational efficiency yeah so I believe it's you know artificial intelligence as you said before is a really good Technologies that's help us you know to process a lot of data in a very in a very short time so and I believe this is the biggest advantage of artificial intelligence as you know human can you know can be tired you know we can lose our Focus but yeah I you know it's just work as a machine so it it it shouldn't be tired maybe could have some hallucination as you know yeah but in general it works very precisely so yeah how do you think it can improve for all these systems in any chance like what systems in particular uh I mean all the computer computer vision decision making you know uh so every single said to me I've discussed before you know when yeah I believe it's you know now it's mostly all the systems are cloud-based because you have to process a lot of data and I believe that somehow it would be later more embedded you know that's how they can become more autonomous but for now yeah probably it's just to to to complex to discuss right now but yeah I believe it's some benefits could be you know this smart assistance you know because for now if you will come back to the topic of ethics it's very important who take decisions and I believe this is the question I was asking in this PhD presentation and I was asking will take responsibility for the decision yeah because AI you know still nowadays the biggest concerns about your eyes it's still not ready to give responsibility to AI to make decision for us and uh if we are talking about responsibility the persons who who make decisions are mostly take responsibility even if they are using AI so I could be a really good assistance but not decision maker so what's your opinion on it yeah well if we look at generative Ai and how how the content writes and the copyright is being kind of evaluated for for the outputs of those of those models it's really a great it's a gray area right so with with respect to decision making responsibility who made that decision you know and then asking the question who made that piece of art you know the the training data for you know the Corpus of text or the Corpus of images um that something like mid-journey would use to to Output a piece of art is it the person that generated The Prompt so nicely or is it all of the artists it's is it a small weighted sum of of attribution of all the artists that contributed that of which their Works were sampled um to create to generate that piece of art by mid-journey you know is it a sum a weighted sum of all of the prior past decisions that were referenced in the model architecture uh during inference time that share the responsibility share the way the responsibility for the decision that was outputted by the model you know was it the model itself was it the person that commissioned the model um was the person that you know commissioned the project itself like you know these are all important questions right um but obviously it would be nonsensical to say okay your training data was in my was used by my algorithm therefore it's your fault that's that doesn't make sense right because they didn't consent to that um but yeah you know it's it's tough because if if mid-journey uses like elements of Van Gogh of of Salvador Dali and all these famous people like and it doesn't know how to kind of pay attribution towards that inspiration I think I think there should be an explainability um the less black boxiness to to the way that these things are generated such that it can point to the influences point to The Inspirations right that that should not be an indictment in in cases of things going wrong but it should be a way of saying why did you make that decision you know because a human if I ask you you know like with respect to your startup like you know why did you do this this particular way you would you would tell me in the past I I you know I ran into this issue I and then that makes me think this way because of what I experienced in the past you know algorithms are not really good models are not designed for that kind of explainability just yet so until until we get to that level of explainability it's really hard to find culprits other than the people that put that algorithm in service yeah yeah right or you know who who curated the data set you know I I'd say um the stakeholders in the machine learning engineering and machine learning Ops you know Cycles you know you have machine learning Engineers that just have to have to decide what architecture you have to decide how it's operated right and then you have to design what data is trained on right you choose the wrong data set for your use case you can have that can have significant consequences right but also you could have great data but a terrible model architecture and then you can have both great data and a great model architecture but the completely wrong use case and so it's a complicated question yeah but but you you know or the [Music] um remind me some some thought yeah there was some posts not so long time ago and in some social media that was a really good example of combining large language models and knowledge graphs that's how we can came to this explainability and I believe this is the right way to go because uh now it's not important on the get the final result but also understand how these results was I mean to the intermediate yeah so I believe this is the right way to go yeah I will Skip One One segment I don't I don't we just already discussed a lot yeah and I I would go to the next step let's talk about startups and showcases uh harvesting Innovations will drive the future of Defense and let's try to feature some some hard tech startups yeah probably your your salad will be also one of them yeah yeah I really wish you and to be one of them and yeah let's exemplify the convergence of drones AI Knowledge Management and computer vision in defense maybe not only in defense so you don't need to be uh you know um [Music] limited on the defense so because you're you're your startup is not directly connected to the defense but yeah it could be used it started in defense yeah and we're focusing on the commercial use cases and then we're going to keep all our government customer Discovery and you know those customers we're gonna say Hey you know we're validating this in the commercial markets but we know this is dual use case Tech which we can we can later bring back to you so okay let's try to figure some some just show some examples and why do you believe your startup will have this great opportunity in the future yeah so if you look at two two examples like stadio and Brink and so skydio that's kind of the American competitor to DJI right and they they required a lot of funding you know over 200 million dollars in consecutive funding rounds so that's when investors see that they're just like holy crap I don't know if I want to get into this like you know it's it's it's a huge funding risk essentially like are we going to be able to con you know because because of how capitalally intensive starting something like scottio is you know there there's a lot of doubt in the succession the rapid and timely succession of funding mm-hmm right because over 200 million dollars that is a very very rare case that is not going to happen to every heart attack startup even if they're doing the right things potentially and then if you look at Brink uh you know another drone American drone company they raise over 80 million in consecutive funding so when you know investors you know they when they hear about cause of twin they're like you know let's let's um let's de-risk the technical technical execution you know maybe can you Outsource um some part of your technology stack and you know like I to me that that is a good idea but that also means less control over you know the entire technology stack right and you know we're we're trying to build a vertically integrated solution uh as much as possible because that that equates to more defensibility right but it also it requires more r d right so the reason why I brought up those two companies is because they they have unique segments in which they service right skydio is they have contracts with the military they have contracts with law enforcement police departments fire fire fighter fighting departments um in certain cases um they they do car crash investigation they do um hostage um scenarios they do um people resisting arrests and car chases and running around in the woods they kind of they provide support in those use cases so and and brick as well but Brink unveil the technology that's actually similar to ours but more so for the use case of indoor hostage situations indoor standoffs right um with with criminals or suspects or finding people Trapped By by like you know hostage taker so they they service law enforcement and Military and you know I'm just like okay this is great they found that Niche they're serving these people and you know the technology is is there um for all of us um but I I just said okay let's let's service Architects let's service Construction Construction managers and let's make our scanning technology a little bit more High Fidelity than what Scotty or Brink can provide because what they provide is good enough for these scenarios where you just need to know just enough about your environment to navigate it right but in the case of you need more information that architecture would require like you know quarter of an inch or an eighth of an inch accuracy or Tolerance on a scan of them right that that is a little bit more involved more technical challenges and that's that's the path that we're going down you know we want to literally recreate buildings as if you know as if you were there in person uh and and essentially the mission of the company is to give people um the tools to prevent themselves from having to go there themselves especially in dangerous scenarios you know and in the case of how we started the company you know entering ballast tanks is very claustrophobic and the atmosphere can can be um polluted High methane concentration in the case of construction sites it's not there there are areas that are not very passable at times you know and preventing foot traffic in theory uh well decrease risk of occupational hazard right so so yeah you know uh we're trying to make ourselves you know when people ask us about the funding risks you know we're different this guy don't break they're doing the law enforcement they're doing Public Safety they're doing military uh and while we're eventually going to do military as well you know the military has a lot of room for for working with the private sector either through sbirs phase one um direct to phase two um and then obviously the the larger larger contracts as well uh no one can hack claim a complete Monopoly on on Military business at least um not with the way that I've seen it operate um yes you can you can the competition can try to Shake You Down buy you out but I'm not trying to compete with skydio or or Brink I I think what causal toy is trying to do is just support a completely different customer base in a completely different way so that's that's kind of how I deflect the criticism um or not the criticism the the concerns that investors may have um so yeah that yeah it seems you you already asked all my next questions because it was exactly about what you were talking about I mean this unique uh you you already highlight of unique products and Technologies so yeah maybe the one thing I just want to to ask you about casualty twins right um so what what are you looking for in the in the next step so uh probably yeah some some people that we will be watching us they can hear you yeah so maybe you can ask about some partnership or whatever maybe even fundraising uh yeah yeah so we're you know we're looking for precede funding and eventually I think uh by the end of this year we're going to cross over into the seat territory and you know it's been bootstrapped so far we've been operating for a year uh but you know we're over the summer we're gonna have the Prototype built uh we will have a pain design partner signed that's the goal and then you know once we de-risk some of the technical execution and also the the business risk I think we'll be in a much better place um for fundraising but you know as with as for anyone when it comes to this it's it's not about uh rushing things right yeah you want to make sure that that you're on to something there's enough interest from from businesses in the case of B2B and consumers in the case of b2c uh before you spend a lot of resources and time money effort and all that so you know while we we do you know we we believe in in the mission and the product it's just we need to get that prototype built uh we need to do some more validation with with you know some of the customers and um and then yeah I think it'll be a very exciting summer for us yeah I sees it's you are a really good student of siblank yeah that's exactly that he is also telling all of their his students yeah so I just can you know edits you always can count on on my company as we are doing some knowledge management software as well and I would like and happy to to to help you and your startup also we have a really great Community around the world and there are mostly the founders business owners experts and also some investors private investors maybe some Representatives yeah so yeah it's nice to you know to to cooperate if it's reasonable and yeah so I will just make some wrapping up of what we already have discussed so yeah we've talked about drones and uh and artificial intelligence it's good and real a revolutionized defense with Knowledge Management so as you could hear all these Advanced Technologies are evolving dramatically it is used almost everywhere and I believe it will be even more complex solution later so Kyle is doing his startup in in uh area of property Tech yeah yeah so you see how many interesting things are happening and that's why we are doing our time garage show so at least we can we can talk and discuss the most interesting events and then and startups and all the Technologies yeah remarkable Technologies I should say that's partnering the hard text startups the software startups and shaping the future of defense or civil um you know operations so making them smart and more efficient and more secures and even before so stay tuned for more exciting Innovations and startup stories Yeah in our upcoming episodes and remembers it's the future size in garage so I would like to thank you Kyle for assist great conversation yeah it was a bit spontaneous I should say but it's very exciting and I'm really happy to have you as our first guest in our stotram garage show so thank you so much we will come back to to our conversation later but we will make some uh progress yeah in stand up so we can at least to check where what's happened and what probably happened next of course yeah thank you yeah so that's it okay

Channel: Startup Garage Show

Views: 2,941

Keywords: sgs, sgshow, startup garage show, livarava, artem kariavka, startup, ai, drones, knowledge management, kyle orciuch, startup business ideas, drones with cameras, artificial intelligence, artificial intelligence news, artificial intelligence robot, knowledge management system, artificial intelligence and machine learning, ai technologies, drones 2023, factors responsible for growth and development, computer vision, hard tech startups, drones in defense, dji drone, openai chatbot

Id: wRsJjI1Nr-c

Length: 71min 30sec (4290 seconds)

Published: Tue Jun 20 2023