전사 🔗
WEBVTT
00:00:19.600 --> 00:00:21.839
Hello and welcome to Full Tech Ahead.
00:00:21.920 --> 00:00:26.000
I am your host, Amanda Rizzani, and I'm excited to have Ophir Mula today.
00:00:26.079 --> 00:00:28.239
He is the CTO of Lumana.
00:00:28.480 --> 00:00:29.280
How are you doing?
00:00:29.839 --> 00:00:31.120
I'm fine, thank Amanda.
00:00:31.280 --> 00:00:31.600
Thank you.
00:00:31.760 --> 00:00:32.799
Thank you for listening me.
00:00:33.119 --> 00:00:35.039
Yes, excited to have you on the show.
00:00:35.280 --> 00:00:37.439
Can you share a little bit about Lumana?
00:00:37.679 --> 00:00:39.439
What services do you provide?
00:00:40.640 --> 00:00:46.799
So, Lumana in general is uh we're doing a lot of stuff with visual intelligence.
00:00:46.960 --> 00:00:57.520
We are coming to businesses that uh invested billions of dollars on their cameras, and we are making the information from the cameras a meaningful information that they they never had.
00:00:57.840 --> 00:01:12.879
Uh, as a starting point, uh we are coming to uh, and it can come from smart cities to uh to uh some retail stores and and businesses that are just you know employees that are working, connecting to the cameras.
00:01:13.040 --> 00:01:18.400
After that, they can put events and insights that they want to get from them.
00:01:18.560 --> 00:01:28.400
It starts from basic physical security, like making sure that no one is uh breaking my business during non-operation time, but it uh it's going much beyond that.
00:01:28.480 --> 00:01:40.640
It's and with you know, how efficient I'm giving a service, okay, or uh should I get more employees to be uh operating in peak hours and how much will be my revenue?
00:01:40.799 --> 00:01:46.959
Uh at the end, these visual these cameras were recording a lot of stuff that no one was using, no one was looking on.
00:01:47.200 --> 00:01:56.640
And we are making it from this and meaningful information that now the business owners and the managers and the employees can be much more efficient using it.
00:01:56.959 --> 00:01:59.519
So taking that data that really was being overlooked.
00:01:59.760 --> 00:02:00.719
Exactly, right?
00:02:00.799 --> 00:02:06.079
Uh till now they they had they went back to the data just if something bad happened.
00:02:06.159 --> 00:02:11.840
Okay, let's say that you know someone was uh stealing from the cashier or or someone was breaking.
00:02:12.000 --> 00:02:17.599
And at the end, there is a limit on how many cameras, how many streams you can use, right?
00:02:17.680 --> 00:02:19.120
Uh or you can watch.
00:02:19.360 --> 00:02:26.080
Even when they were control were connected to control centers where there were people that were kind of watching the cameras.
00:02:26.400 --> 00:02:29.759
Uh, there were researchers that after 10 minutes they got blind.
00:02:30.000 --> 00:02:45.039
We are having AI agents that are operating on each one of the cameras, collecting insights, making sure that nothing bad happens and giving meaningful information that you can do something with it and being more proactive and less reactive to something bad that happened.
00:02:45.439 --> 00:02:48.800
Okay, we're gonna talk more about video surveillance today.
00:02:48.879 --> 00:02:53.039
And my first question, of course, this is the big topic AI.
00:02:53.439 --> 00:02:56.879
So, how is AI changing this field?
00:02:57.840 --> 00:03:00.080
Wow, in so many aspects.
00:03:00.319 --> 00:03:08.400
At the end, you can think about it that right now on each camera, I can put similar to human.
00:03:08.479 --> 00:03:18.080
Uh, what if you could put someone and you can give it a task and and and just ask for it to get your notification based on what you are asking?
00:03:18.240 --> 00:03:23.439
And you can ask it, you know, whether you're a sing person in a certain time, just notify me.
00:03:23.599 --> 00:03:32.400
But you can ask it also, and you can tell this uh someone, this agent that is looking on your camera, summarize to me what do you think about how the operation looks like?
00:03:32.560 --> 00:03:38.479
Or what do you think, what is something bad happened, or is something unusual, even without defining it.
00:03:38.800 --> 00:04:00.159
Because when an agent is sitting on your uh on your camera, it's kind of learning it, it knows what is usual, what is unusual, and and you can get notified and get insights only for the cases that matters, instead, and instead of having a huge bank of data, just get a small insights of whatever whatever is happening on the camera.
00:04:00.879 --> 00:04:01.199
Okay.
00:04:01.919 --> 00:04:05.680
So a lot of companies are operating with cameras they already have.
00:04:05.840 --> 00:04:12.080
Is there a way that Lumana can bring these new capabilities to the cameras they already have?
00:04:12.960 --> 00:04:16.800
Yes, uh, Lumana is actually we are not bundled to certain cameras.
00:04:16.959 --> 00:04:30.639
The best thing about Lumana is that all we need is you to give us the permission to put uh our compute in your network, keeping the data on your site, meaning that you know we are even by that we are keeping privacy.
00:04:30.720 --> 00:04:36.800
You don't need to upload a lot of video data to the cloud, you don't need to connect it to some you know expensive services.
00:04:37.040 --> 00:04:48.959
We are putting our compute, our AI pipeline in your network with all the with your data, no matter which camera that you are giving to us, and you are getting starting to get the proactive insights from it.
00:04:49.519 --> 00:04:50.319
Interesting.
00:04:50.639 --> 00:04:55.360
Can you share any real-world examples from a company?
00:04:55.439 --> 00:05:01.839
You don't have to name them, but just where they caught information they may have missed before and how that uh impacted the company.
00:05:02.160 --> 00:05:02.879
We have so many.
00:05:03.040 --> 00:05:07.279
I'll give you one, I'll give you a few on the physical security side, right?
Load full transcript
00:05:07.439 --> 00:05:12.319
Uh imagine we have a large city that is a customer of us, right?
00:05:12.639 --> 00:05:16.959
And and and things happen in, you know, uh a kid got lost.
00:05:17.199 --> 00:05:18.639
Okay, really got lost.
00:05:18.800 --> 00:05:28.319
Now let's let's search for a kid in uh all the cities' cameras and start to track from which state from which street he went to which street.
00:05:28.480 --> 00:05:43.920
In seconds, Lumana gave all the information, what is the latest place that the kids were looking on, the fact that it was taken to a bus and the police got involved, and after less than an hour, the kid was safe in his house, in his parents' house.
00:05:44.079 --> 00:05:46.240
But it's not just physical security.
00:05:46.399 --> 00:05:48.639
Uh, we came to a certain business.
00:05:49.120 --> 00:05:55.839
They had some kind of a compliance policy to make sure that things are operated based on what they defined.
00:05:56.000 --> 00:05:59.920
When it was in this task, it was how you are cleaning the floor.
00:06:00.160 --> 00:06:00.480
Okay.
00:06:00.879 --> 00:06:09.360
And there were supposed to be managers that were kind of had to notify yes, uh, the floor was cleaned at least for 10 minutes.
00:06:09.439 --> 00:06:13.920
Uh, the employee was wearing gloves, uh, he was not talking over the phone.
00:06:14.079 --> 00:06:16.879
And and for that, he had to come to interview him.
00:06:17.040 --> 00:06:21.279
From time to time, it looks on the camera to make sure that he's doing his operation properly.
00:06:21.600 --> 00:06:25.759
But what Lumana is having, we have the agents that are kind of looking.
00:06:25.920 --> 00:06:34.800
We know how to identify what is an employee, what is the operation that he's doing, classify how much time he's doing, making sure that this time is continuous.
00:06:34.959 --> 00:06:54.639
And instead of doing and involving humans to do a task that is just, you know, marking checkbox to make sure that the work is done properly, using Lumana, you get also the physical security, but you get also insights for your business, something that is reducing dramatically the cost of the operation that you need to handle.
00:06:55.040 --> 00:07:00.240
Okay, so that brings another question to my mind is you're using these AI agents.
00:07:00.480 --> 00:07:06.399
How do you ensure that it's accurate information and also the useful information that they need?
00:07:07.680 --> 00:07:09.199
Uh, it's a very good question.
00:07:09.439 --> 00:07:14.560
One thing that is unique about Lumana, and you can check me, no one in the industry is doing it.
00:07:14.720 --> 00:07:21.759
Uh, Lumana is the only solution that is using uh self-supervised learning on each one of the cameras.
00:07:22.000 --> 00:07:27.759
We believe that the camera that is now, let's say that you place camera on Miami Beach, right?
00:07:28.160 --> 00:07:38.800
It doesn't look people are walking dressed differently in a different angle, and it doesn't look like people, you know, in uh in in Alaska during the winter, right?
00:07:38.959 --> 00:07:42.959
You need to teach the camera based on the camera visual content.
00:07:43.040 --> 00:07:46.079
You need to teach the camera based on what it sees.
00:07:46.319 --> 00:08:02.959
Lumana is the only solution that is pushing a unique model, a unique agent per camera with a self-supervised learning, meaning that there is no person involved, no one is looking on your camera, and you are getting a model that is tuned to what this camera is focused on.
00:08:03.120 --> 00:08:12.639
And if you are putting our solution in a pathing log, you are getting a model that is optimized to how vehicles look like, what is the typical operation.
00:08:12.800 --> 00:08:29.279
And if you are putting us on Miami Beach, right, we are more focused about people, we are more focused on and we are giving you a novel accuracy and also tracking your accuracy over time, meaning that if something dramatically happened, let's say that uh winter is coming, right?
00:08:29.360 --> 00:08:40.159
And and people are dressed differently, the model is keep being updated and keep being tracked with its accuracy and making sure that all the time we are kind of staying on the same level.
00:08:40.320 --> 00:08:47.279
So you're getting a much more reliable information from uh the AI that is using, because I agree, right?
00:08:47.360 --> 00:08:54.159
If you're putting some kind of insights and suddenly you're getting thousands of false alerts, you don't believe in the product.
00:08:54.320 --> 00:08:58.240
You need to make sure that the accuracy is maintained in the novel accuracy all the time.
00:08:58.559 --> 00:08:58.879
Okay.
00:08:59.679 --> 00:09:08.080
One thing I know people are more worried than ever these days about their own personal privacy in this digital AI age.
00:09:08.399 --> 00:09:18.799
So, with all this new camera surveillance, how do you ensure their personal privacy is met despite all the surveillance?
00:09:19.600 --> 00:09:24.399
Privacy and security are the kind of ranked in the first level that we need to meet.
00:09:24.480 --> 00:09:29.360
Because if someone is getting access to your data, we basically we didn't do anything.
00:09:29.519 --> 00:09:33.360
So, first, our training is a self-supervised approach.
00:09:33.519 --> 00:09:34.960
No human is looking.
00:09:35.120 --> 00:09:46.159
Whenever on your data, so when your data is getting trained, it's getting deleted immediately, all the data that is used, in despite the fact that only machines are kind of the one that is looking.
00:09:46.320 --> 00:09:47.600
This is a first level.
00:09:47.840 --> 00:09:50.960
The second one, we are a really true hybrid solution.
00:09:51.120 --> 00:10:09.759
If you want to keep the data located on your network, the visual, the video, or the images, or even our training engines, we are giving you a solution that you can include everything inside of your data
centers or in your uh object storage and uploading only metadata information.
00:10:09.919 --> 00:10:23.200
The data on our cloud is secured, and we are using the uh the best kind of in-class security aspects, but you are having the rights to choose the proper architecture to your solution based on your needs.
00:10:24.080 --> 00:10:31.600
Well, looking toward the future and how rapidly technology is advancing, where do you see AI surveillance going next?
00:10:31.840 --> 00:10:33.039
So, how did we start it?
00:10:33.120 --> 00:10:35.840
We started by people looking on the visual content.
00:10:36.159 --> 00:10:42.799
And then what happens, and it's now let's say we are in this phase, we I think that we're even beyond it.
00:10:42.879 --> 00:10:45.759
Uh, we started to tell AI what to look for, right?
00:10:46.000 --> 00:10:53.039
Let me know if someone is coming on a bad time, or check if this person uh is arriving because I want to get notified.
00:10:53.200 --> 00:11:09.200
I think that the next level, uh, and this is where we are going to, is that you will connect the cameras to your AI agent, they will learn how you operate, they will learn what is the typical environment condition, and they will just let you know when something bad happens.
00:11:09.440 --> 00:11:13.919
I don't need to, I don't need to specify what is a bad thing, right?
00:11:14.080 --> 00:11:18.080
I have my usual behavior and I have my unusual behavior, right?
00:11:18.240 --> 00:11:23.519
I know what is my operation, and then they can tell me how to get better and better.
00:11:23.679 --> 00:11:32.639
So I think that at the end, same thing that as we got and we gave to the LLMs and information on the internet, and they kind of uh teach themselves, right?
00:11:32.720 --> 00:11:56.960
As long as we will give access to this visual content, making it secured, privacy, uh, which is the highest priority from our perspective, and and and and building agents that be able to learn and teach themselves on what is usual, what is unusual, we will be able to get much more proactive insights without having human the need to specify what is unusual, right?
00:11:57.200 --> 00:12:10.879
Because even if you will now your only job will be to sit in kind of the uh on on uh on the 5th street in uh in New York, and you will keep continue looking on the street 24-7, right?
00:12:11.440 --> 00:12:17.679
At the end, you will be the best expert to know what is usual, what is unusual, and what should get notified for the police.
00:12:17.840 --> 00:12:25.279
There is no there is no reason the AI agents over overcome us will not be able to give you the same type of information.
00:12:25.919 --> 00:12:26.879
Absolutely.
00:12:27.120 --> 00:12:37.120
Well, if business leaders want to harness this new video technology for their company, what questions should they ask and what is a good first step for them?
00:12:38.320 --> 00:12:46.320
If they want to address it, they need first to understand how much this is able to be adaptive to my business.
00:12:46.720 --> 00:12:51.919
Okay, because physical security is basic, right?
00:12:52.159 --> 00:13:03.120
Um being able to notify that the person arrived on a certain time, this is something that even you know done today on the camera, it's not enough, right?
00:13:03.360 --> 00:13:09.279
I need to understand how much you can adapt your solution into my needs, right?
00:13:09.840 --> 00:13:27.360
Because if my needs are to make sure that uh let's say that you know uh I'm I'm I'm I'm a business that uh is now doing pickups, okay, and I need to know whether my uh employees are working properly, you need to identify very fast how a pickup looks like, right?
00:13:27.600 --> 00:13:29.759
Or if my business is a bank, right?
00:13:29.840 --> 00:13:36.480
And I need to uh make sure that the lines are short, you need to be able to give this me to give me this information.
00:13:36.639 --> 00:13:41.679
I don't need to spend now hours or days of training, right?
00:13:41.840 --> 00:13:51.200
Your solution needs to adapt itself to my needs, and I think this is the main the main question that customers need to ask.
00:13:51.519 --> 00:13:58.000
This will tell them whether the solution that they selected is scalable and they can get information that cares for them.
00:13:58.879 --> 00:13:59.200
Okay.
00:13:59.759 --> 00:14:04.480
Well, if there was one key takeaway you could leave our audience with today, what would that be?
00:14:05.840 --> 00:14:16.320
I think uh they need to start to understand and ask what are the insights that they are expecting to get from someone that is looking on their space.
00:14:16.480 --> 00:14:18.559
And imagine what they can get, right?
00:14:18.639 --> 00:14:24.480
If they put humans that are just looking 24-7 and what are the questions they will ask them.
00:14:24.799 --> 00:14:41.840
And this is what they need to expect from the solution that they are putting on top of the cameras, because the information there is a valuable, and only a solution that is hybrid, scalable, adapted to their business will give this this type of information.
00:14:42.320 --> 00:14:42.879
Wonderful.
00:14:43.039 --> 00:14:46.559
Well, thank you so much for coming on the show and sharing your insights with us today.
00:14:47.120 --> 00:14:48.799
Thank you, and have a nice day.
00:14:49.200 --> 00:14:50.799
You too, and thank you to our audience.
00:14:50.879 --> 00:14:52.879
If you have any questions or comments, leave them below.
00:14:52.960 --> 00:14:54.960
I'll try to respond back as soon as possible.
00:14:55.200 --> 00:14:56.799
Have a wonderful day.