🎉😎 125th SPECIALL | ROUND TABLE #1 | RAW TALKS WITH VK

0:02 New attempt. >> First ever round table on Roblox. >> This is very interesting. >> Can see one second. >> Hello. Hi. So, >> wow. This was impressive. >> This is my problem. >> Can we do all of that? >> Yeah, let us try. We'll start. But is that possible? And then but it's too [ __ ] risky. Okay, let's get to the answer. Yeah. Yeah. Okay. >> That is insane, man. prompting clarifying what you want. similar

0:48 Gemini 3.1. What are those crazy things that you do with AI in your system that you won't advise anyone to do? >> What are those five top channels that you follow? Video is AI. My audio is AI. My thumbnail is AI. How much are you making from YouTube? Less than using. And it's a part of evolution. If you want the wheel to go, then the square has to cut off his edges. Girl.

1:33 >> Is this the first time in your podcast that we're eating? Actually, >> no. But you know what? I had a lot of fun. Welcome to the show. >> Thank you. >> One eternity later. Okay. So new I should I need a wish me good luck and

2:25 we have done open she's all super active in front of the camera I don't know what happened but anyways thanks thanks for accepting finally. >> Welcome to the show actually. So after yesterday's conversation, I was thinking why am I just daring to just stay away from certain things. it's time for me to actually explore properly especially adapt I seriously wanted to push that but yeah so today uh completely topics but in a practical way let's

3:12 brainstorm we'll rope in few guests also surprise element so where we're going to take their problem statements let's explore I Before getting into the proper crux of the conversation prompt prompt engineering or prompting word I used to feelen over celebrated but with time every month every every activity I do on platforms chatboard I understand I keep getting the uh importance of it important like for example she wasn't fast so

3:58 She was she was complaining headache headache. Okay fine. So she opened common she told this is what is happening since yesterday headache and I on today. This is what I know. Now ask me more questions to understand my situation. This was very new for me. Oh, this is a new new way of asking a question to a chatbot. That means it'll it'll give a list of option. It gave list of options and with patience that was single 30 long questions. She gave it almost

4:46 good in half an hour or 45 minutes time. So utilizing an leveraging leveraging an AI to actually understand how to use use an AI now I understand why is it so overspoken why why so yeah what do you think what's happening around around us yeah please welcome welcome to the show thank you thank you coming to prompting so prompting It is always clarifying what you want. Okay.

5:35 But I I'll use it again. Okay. >> Okay. So ambiguity clarity prompting similar he was mentioning. Okay sorry.

6:28 You know what is happening? I don't know if you've heard of context engineering. Conting. What is context engineering is exactly what we are in a way telling and what you guys are doing also on matter. Okay. >> Prompting prompt engineering is question that is prompt engineering. Context engine. How do you give the AI model enough context for it to be able to solve the problem? Context mostly the moment you go through the adapt framework and go deeper by default we are becoming context engineers >> right if you see my prompts and all right okay I want you to do this task for that you

7:13 need to go to my email for using my email use the Gmail MCP >> right >> tool context system. The other thing I want you to go and fetch the data of my YouTube analytics. For YouTube analytics, use the Chrome browser using browser views. Browser views tool which AI has access to. >> But you're not letting AI to figure it out itself. >> The context is to know. So it's context is all about making it easier for AI >> to understand what it needs to do. >> Right? So in this both of you agency framework

7:58 still you you'll be using this can you know sure I think uh again right dabbling we keep going back so I use everything okay >> and I keep using everything the reason is it's our job to teach everyone also what is working >> right genuinely go to >> uh now back of my head is based on use case >> right >> go to for example deep research about any of the topic right now my answer every tool has deep research but I use perplexity majorly >> feature it's called model council

8:46 >> what is the power of chad GPT it's model chat GP model and GPT >> Gemini is a model is Gemini anthropic model CLA's model or anthropics models are opus and so on using something called as model console okay so is going to show you very quickly what model console is >> absolutely Okay. Right. >> Okay. So any business idea that you have, you want it validated. >> Okay. So basically it's majorly for um optimizing retail outlets. So we have a

9:31 lot of uh you know especially when we talk about uh restaurants there will be a lot of people coming in and sometimes we don't address those people. So there was one business idea they were completely into computer vision CC cameras. CC cameras they've connected something. I'm not sure what they connected. So it computer vision technology it captures everyone's face and a table and let's say if someone is not addressing them it'll ring their bell inside and that person will go take care of them how much time are we wasting because sometimes we keep talking especially when we go to cafes we keep talking talking a lot. So end time. So to optimally utilize all of these things uh in Chennai has started doing that.

10:17 >> So that's the business idea. >> Okay. >> So this is whisper flow. Whisper flow. I think last time. Yeah. So this is a tool that we use. In fact everyone in our company uses to convert voice to text. There was this video also they said they'll give a Ferrari if we'll give you a right. I've never tried this. Kudos to a good friend. So very good friend for the product. I didn't realize that there

11:09 in future they will make it to other versions also for sure eventually they release it to others. models GP 5.4 thinking model. Okay. Gemini 3.1 Pro. Nice. Okay. Okay.

11:57 Nice. So basically it's not like so there's a discussion. It's not like it's like different versions. This is like different, you know, AI models discussing together and then coming up with an output. >> Crazy. >> Yeah. Yeah. Go for go for it. >> So, council GPT 5.4 is thinking problem statement. zoom in a little.

12:43 So basically what is happening is okay free model Google the free version chat GP free version cloud free version open you can take it to all the three models I'll tell you what it's doing three different AI models each of the AI model let's say chat GP Gemini claw answer Gemini chat GP in whatever the other mod so we giving answers of other models also to each one of the models

13:29 then basically you are simulating that and eventually report on a high Okay, nice. So, ohc business idea validation computer startup retail optimization startup ideation preferences. Okay. Okay. AI product business validation found. Okay.

14:25 So as it's happening there is something called as gro heavy. Okay. what he said a debate debate usually happens in one layer. In fact, I think Kimmy also released a similar version as well. Okay. So problem statement instead of that is the that is how AI work every AI works differently right then gro heavy learning it will spin up multiple instances of same gro it's like okay GP

15:15 it will take different direction every time right >> so gro heavy does that for you >> so multiple instances open by default because hand off it will make it forcefully think in different ways. >> Mhm. >> And eventually debate with each other and coincide with a solution. So but model advantage problem statement validation with different AI models there it's the same AI model competing with each other problem and validated

16:04 do not use facial recognition use anonymized tracking there's DPDP act existing CC use existing CCTV don't sell hardware hardware already solved problem software Yeah. Target organized restaurant chains first, not single outlet confidence level

16:55 but sharp strategy viable and timely. Okay. So market because that it didn't get the starting city comprehend The whole final recommendation idea worth pursuing idea worth pursuing but only with a

17:41 privacy first anonymous approach >> targeted at Indian restaurant chain solving the specific problem of customer attention latency and table turnover. >> Don't build generic restaurant AI. >> Build the system that ensures no customer goes unattended. Right. Recommendation problem statement definitely worth paying for it. That's what I I personally I didn't know this

18:26 unfortunately. could build it because publicity has all the models together. >> What I like about Perplexity Model Council is every model has its bias. Every model works differently. It's like five smart people are sharper in one area. Yeah, unique. >> Unique in their own ways. I'm able to get five smart people to work for me rather than one kind of smart person working for me. That's the advantage. >> So deep research key you're using publicity. >> Correct. >> What are those other key areas or key?

19:11 >> So let's make it very quick. In fact, I might have a different stack. He might have a different stack. >> And that is the beauty major for content research and all. It has access to you know Twitter data. So content research right. So Grock has access to all the tweet data which it can pull. So which is great but content rewriting >> anything creative >> okay >> creative writing because Gemini 3.1 is very good with creative writing better than everything else based on my taste >> that's the point of anything on the go

19:58 very quick email for anything development coding or anything like that I usually use either claude or I will go to something like miniax 2.7 >> m >> right miniax 2.7 Chinese model the you have to use something via open router but okay >> claude is basically very good and imagine 5.4 for codeex is also very very good which is by open that is with coding now you'll be like developer >> you will see over the course of today >> problem statement it doesn't matter if you're a developer or not all of us are building >> right >> right so selecting the right model is very very important

20:44 anything with automation agent build it's majorly cloud models So that is my high level stack if I can imagine >> and context you were mentioning about couple of Chinese models or tools say for suppose I'm a I'm a guy who's against of these Chinese apps but I should log into that tools what is the solution there you if you use a tool like open router if I can share my screen Chinese because you think it's not safe and all safe and all first of all

21:37 studio is one where you can install a app like chat GPT or you also have something called as okay same high level download button. Why is it a download button? Because download first of all it will download that model. Look at this AI model which is open source. >> Okay. >> Right. GPD anthropic models closed source and there are lot of AI models by open most of the Chinese models are open source. >> Okay. 10 3.5 open source and all openware

22:34 I can install an AI model on my computer >> which is happening right now even if you don't have internet even if you're basically traveling on a on a flight or anything, you can use this AI model to to run. >> So, every computer has a GPU >> computer, Windows, Mac, there is RAM, then there is graphics, Nvidia GPU and every computer has GPU. So contained. It's already trained. Okay. Got it. So

23:21 GP 3.5 4.5 with no one. How are they answering? Because all these models are trained with a lot of information unlike humans. All right. So it's able to answer it's okay >> but you should not use it for you know things like uh how do I put it facts

24:07 like Hyderabad ai because that needs recency based information >> right >> which these models don't have internet I can add that internet access >> okay >> then it will be able to there you go hello how are you >> u there is this uh creator named Wamshi and I want him to invite me on his podcast so write a super convincing email uh to which can convince him to get me on his podcast. He's a creator in Telu. Uh use the information if you have any about him.

24:56 >> By mistake four is the new model basically >> but scroll. >> So look at this. >> It's taking some assumptions right now. And I'll tell you the beauty of the answer that we getting. It's coming from an AI model which is 12 billion parameters. >> 12 billion I think it'll be a trillion 1 trillion but >> I think it's close to a trillion parameter. So this is 200 times less powerful than an opus >> but it's still able to do most of the work that you want to do. It figured out that you're a person >> crazy >> and it came up with this email. It's fine. Let's say you have a thought

25:42 that this can do very well but don't give it fact information. How does that work? >> How does that work? It'll just create a temporary email server for you. >> Okay. >> Uh it's called burner email. Mhm. >> Uh by the way, LM Studio is also one more similar app that you can use, >> right? And for phone, Google AI edge gallery again are they same opensource >> open weights model. You want AI to run without internet anything install it on the phone. Now >> injubo it was a 12 billion parameter

26:29 model, >> right? >> But on your phone 12 billion parameter no >> it's too big. So it'll be like 1 billion parameter but still is very good trust me when I say this. >> Okay. >> So what was it burner uh thing >> next? Uh basically data this is AI >> you can install this same actually impressive >> full UI is like full chat GP style right and the other thing that locally AI does if I'm not wrong is it can run this model. Okay. Apple started to work on a model uh Apple foundation model >> which is optimized for better performance on Apple iPhones. Got it. >> Right. Anyways, so this is locally AI. Next, if you want I don't know if do you

27:16 use a burner email? So, so there is something called so this is burner emails. Let's say platform email listener >> you don't want to be bothered by spam that is coming in >> so instead of giving your personal email burner burner email so there are a bunch of them temp email this is the one that I use which is called as burner emails it's a chrome extension on matter Okay. Burnermail.io. >> Okay. >> This is the one that I use

28:02 once you have logged in every time you click. Okay. >> And because you're using it inside of a browser next Oh. Mhm. >> So you can login with that without actually sharing your email. >> Nice. >> Just in case if your email is still private. >> Still private. >> What are those crazy things that you do with AI in your laptop or in your system that you won't advise anyone to do? Italy.

29:01 System shut down. that is but yeah I think uh AI has access to everything of mine >> literally everything including my stock market data to my bank balances to social media accounts >> to everything so for some time on open

29:49 or you can run it on cloud usually. Mhm. >> Because they're too scared that something will happen in the computer because open claw is like a AI employee. >> Okay. >> Right. So for an AI employee to work very well. >> So there it can do literally anything I want it to do and I can message it on my Slack. >> Now Jerry is built on OpenClaw. So Jerry version 2.0. >> Okay. >> It's built on OpenClaw. So,

30:35 instagram 7day summaries, one day summary, 3day summaries. So, I wish I could show you those, but it actually has sensitive information, so I can't show that.

31:23 My god. Tuesday, first week of Tuesday is usually low. So I'll be like my revenue seems to be very low today. Can you look up if the revenue is accurate? I mean is it actually low today or am I feeling it? should look up all the last 3 months data, last 6 months data because it has access to our revenues data analysis. So what we have done is open we have given it access to all the sources data is connected to something called as Julius. Okay, >> Julius AI has access to the databases of my computer of my company.

32:09 >> Okay, >> it can only read it cannot write. Okay. Already access phone. 24 hours. So open this is context right open

32:55 it name is Jerry again company information so it's like an employee for Right. Chief of staff or whatever like it's it's what it is. >> Interesting. >> Can can we just see that and >> there is nothing. It's a chatbot. >> It's on my Slack. So that is the problem right with this. It's a SL. It's a chatbot. I remember. So chat

33:44 and also sadly I wish I could show you this but there's a lot of sensitive information. >> I don't care what the technicalities are. We should build something in this video. >> Yeah. >> I'm going to ask something. She has prepared something. Maybe over to you. >> Let's go. >> Okay. So >> ask one of the tasks that you want to solve. We can build that. I think one generic task for is we always want to keep ourselves updated with news. So how can we ensure that multiple because you guys also do a lot of research. So multiple sources information how can we consolidate like all of that information into one single place and three hours so that I can go check.

34:32 Yeah. So I can I can just go check and important thing WhatsApp let's say how can we do that thanks I think we can do this multiple approaches and how much time does it take depends on what you want to do chicken biryani biryani something something because let's say automation WhatsApp you guys said it's a you know aentic process it would take time something

35:18 that a regular audience can use for everyday thing >> not very it is very simple now here is how I break The problem. What is the problem statement? Every last go to you know GPT perlexity. >> What happened? >> Correct. Let's say you go to chat GPT or perplexity or any of the tools for that matter. Let's say in this example

36:04 JSON let's say every of these tools chatb has something called a scheduled task okay and in fact if you have your phone push notification. We can do that first on Claude. Why don't you show? Okay. WhatsApp.

36:53 I will ensure that. Okay. Proof of concept. Okay. cover. So basic

37:43 it can execute things and very importantly system. Okay. Okay. So completely new. Okay. Download every project. Okay. Okay. create your I'll call it >> update AI

38:30 create open. Okay. So, correct. Look up the most uh the look up the AI updates in the last 3 hours and make it relevant to a content creator who is working in the infotainment who is trying to educate people in AI so that this person can look at all of this and could help him in his own work as well as help him educate a lot of people. uh uh try to put a score to this AI

39:18 update based on this and uh put it all in a neat order is like which is for more relevant always a good idea to give a scoring okay and let's So right handress right hand side it will

40:04 Step one, step two, step three. Okay, the more specific the better. The more specific the better, but we are not doing any of it right now. >> Okay. >> So, search for the latest AI news from today. research on major company updates. Second, research viral AI tools and trending topics India specific AI news. So basically automatically break down

40:51 that is what it is doing. Yeah. Whatever is done, it's treating you like a good employee. Nice.

41:37 Great. Nice. LM robotics week different. It is trying to figure out content creator in the enterainment every content angle how you can use it in your own work and how you can explain it to your audience. >> Nice. I will be surprised if the answer is great because

42:24 usually you have to refine keep giving it feedback. It keeps getting better. Got it. It'll be a mediocre. Got it. Okay.

43:25 Google Vids now lets anyone create AI videos for free. Ultra 2 million token context. Mhm. 1 million 2 million 94.3% of graduate level science. >> Okay. Anthropy conveys content creator angle. This is a perfect story for infotainment. Has drama stakes and a clear narrative. >> Let's schedule this. No. Okay. So

44:13 schedule. Okay. Schedule >> this to run for >> every 3 hours. >> Every 3 hours. So advanced like you can run or something. So

45:05 okay done. So first level is something like this. Second level is using with and then automating. Okay. sources xxx every 3 hours whatever schedule we put bring it on WhatsApp correct >> third is the more advanced thing what you do >> yeah but uh the see the one that I do is slightly different I used to use a process like this before >> okay >> what I want is perspectives Google basically but we have sources from Instagram also we have sources from

45:51 LinkedIn also. We have sources from other platforms also and also sources from Twitter. >> Correct. >> But my scraper one of the most important thing that I wanted is every basically Twitter feed the best AI updates feed for me because my Twitter is active on consuming AI content. Twitter feed it is not possible. There is no API which can see my feed. AI can only search based on keywords.

46:41 Because those are perspectives, those are real updates >> and those are customized by default. I need to give my AI >> the power of using my Twitter account. >> Right. >> So every 3 hours, Jerry, my EA, will open my Twitter on my computer. >> Mhm. scroll, pick up the updates and send that to me on my Slacker. What is happening? There's one more screen on the side. What is happening? What is happening? I had to explain them. So,

47:26 so now that this is done, >> I can show you my Jerry. Mine is connected to Slack directly >> and this is my executive assistant that we spoke about. Please blur this corner. >> Yeah. >> Right. Dean I basically you can see top items worth uh right now and non-duplicate because I don't want duplicate ones. >> So update why each one matters. What are the sources every 3 hours? >> Okay. >> This is coming directly from my Twitter feed. But the way it is happening is what makes it powerful. >> Right? >> The news that you will find here, you cannot find it from the internet because

48:12 internet is usually one to two days late. before anyone else talk about it. So I have to pick information from Twitter. I have to pick information from Reddit, >> right? >> And I have to pick sources where other people can't use. For example Reddit API access. >> So I have to do something like this >> or I have to use a web agent, browser agent to be able to do tasks. So this is a slightly more advanced one. This is all done again using uh open rights.

49:10 Okay. And also at least if we are able to tell them tools that go on to make that thing tools if you're able to tell them better. So comet browser okay normal comt browser I'll tell you what comet browser is googlet is like a browser like Google chromets browser is an AIT right for you on your behalf.

49:55 So browser with an AI layer on top of it. Weirdly enough 41 minutes ago web has tweeted while web was sitting in Okay. Now, correct. The only way is my agent has to open the browser and scroll. Correct. Right. So,

50:41 agent scrolling then demonstration is done. You could assistant OpenJC, scroll my Twitter and give me top tweets on my feed on AI that you see. Okay. >> Okay. And Okay. Anyone can do this.

51:30 Nice. So basically I don't have the time to open and say do this task computer insane. On that note and now that you you know get your keep yourself updated about a lot of uh eay what are those five top channels that you follow profiles channels >> something Twitter I mean all of these updates >> h interesting first channel is web hosti

52:20 just kidding uh for YouTube uh there are a few channels that I like watching one is Matthew Berman >> technical it's very Uh two is Andre Karpathi, >> right? Who is the ex OpenAI member? >> Okay. >> Lex Friedman podcast are one more that I'm always on top. Uh imagin I forgot the name. What? TB >> TVPN. >> TBPN podcast which is a daily AI news channel. >> A >> right. It just got acquired by OpenAI. >> Open AI. >> Right. TBPN media company. That is also something. They do a lot of podcast and everything. The other thing is something called as

53:06 the moonshot podcast. Uh that is also a very good podcast. Uh weekly three times usually in Jim >> three four incredible people uh sit and jam on AI. >> Okay. >> Right. Uh yeah, these are the ones and of course Oh yeah, that is true. She said the ones that you follow what I follow and we always have this what channels would they follow? someone that we follow.

53:53 Now that news students who are you know job searched they have to go to multiple then they have to apply. Can we automate all of this process because I heard that like vibe of telling that agent forms they can do that. So how can we do it? So okay again software is it application or >> application but it's like man with a lot of tools. Okay. So

54:44 super concept for job search. So okay. So just normal English. Okay. Go for it. >> Okay. uh maybe you know searching jobs

55:31 for content writing especially for um media agencies >> and YouTube I mean current YouTube channel media agencies. >> Yeah. >> Okay. Which country? Okay. So just country India >> remote preferred >> remote preferred best. Okay. Also include what is the expected CTC? what are the skills especially that this job requires so that at a glance I can

56:18 see what is it that I'm looking for and give me also uh you know some kind of a percentage score on what things information I have given how much it matches >> done also can we also make like some old companies not like startups funded startups maybe >> not bootstrap startups >> uh try to find jobs only in funded startups and not bootstrapped ones >> or large companies. >> Yeah.

57:03 YouTube channel India with remote preference focusing on funded startups and large companies. So any search for content writing jobs in media. Okay, I'll choose for example search content writer jobs funded startups. So

57:50 sheets. Okay. Okay. So it is Google sheets and sheets on steroids. Okay. It could also go and it can look for forms and fill the form on your behalf and attach your resume and apply. Okay. So,

58:37 TVF and you see job title content creator content ops executive content writer intern. Okay. Company type funded startup funded large company funed large company agency funded. Oh, funding status. Location Mumbai, Bangalore, Hyderabad. Remote hybrid hybrid hybrid on-site remote remote remote hybrid onsite. Okay. Doing a summary. >> Summary. Then key skills required

59:24 in case you want to trace it back. You can take the company name and the source and double check. Nice. Nice. Okay. Another easy of double checking. Nice. Okay. CTC breakdown. So this is how it is. Yeah. So

1:00:12 basic version simplest version. >> Simplest version. But even though the simplest version always start Okay. URL

1:00:58 interview story bank it'll take different stories five to 10 master stories that can answer any behavioral question so pipeline integrity for example batch processing Mhm. agent one. So negotiation scripts very important geographic discount push

1:01:44 back and customized for every company with your role how it should do. >> Nice. So basically you need to know what cloud code is. You have to learn cloud code basic. The reason why I'm recommending not to show is one who are not engineers okay that doesn't mean that you have to be a software engineer Okay.

1:02:38 Test. Figure it out. Hey, this is this the backtory of this is this the guy who built this basically jobs. So he built this software so that AI can apply jobs for him. 700 jobs. He got the head of applied AI job also because of this. Nice then more people can use it.

1:03:25 Let's try to brainstorm on problem statements. context. So the surprise element is we're going to open you know seven seven guests from our previous podcast. Interesting. >> So I thought last we'll try to pick few problem statements from students. We've identified different industries different problem statements founders doctor I mean hospital B2B business

1:04:11 problem statement which can be automated through AI. So that might be a question or too much complex situation they waiting already. Let's try to take the calls. I'll not give the list also. >> I'll give you a heads up. Okay. Yeah. >> Right now I'll be very open and frank problem statements and that two from different industries. We are jumping a stage where we cannot talk like the way we were so far. Right now my assumption is me and Dip will talk the operator language or the AI orchestrator language the AI

1:04:58 generalist language right >> yeah so before we get into this quick caution uh questions we will answer or discuss I think we'll have to discuss >> because there's no right answer to a question >> of course yeah >> we have to brainstorm to understand what could be the solution and that will open up a different side of us with some workflows with tools and stuff like that then language is much higher >> because we are jumping from being

1:05:43 dabblers directly to an orchestrator. >> Right. >> You cannot get to the position where you can talk like this. >> Absolutely. >> Right. So heads up. So don't get carried away and if you guys have questions feel free to ask as well. >> Sure. So we'll be laymans or maybe students like that and then complex NBC directly ask you guys and I don't think we'll be able to build everything >> but we can give direction. >> Yeah. reference point not necessarily a building something >> brainstorming on how can we solve problem so even I I'll understand the viewer who's watching they'll also understand even the guest what what would be the process ideal

1:06:28 process sounds good then we'll start so first okay so he's a founder of marketing agency digital marketing agency 100 plus 120 plus >> 120 Yeah. So, friendology would I keep talking to him. So, I thought he'll be the right person also to start this kind of conversations. >> Sure. >> So, uh hi hello. >> So, WBO is here, Pranavi is here, uh Bilipana is here. So, we are starting this conversation. Maybe we'll start with what what do you do uh with your team? I mean with the AI in your team,

1:07:14 how do you utilize or uh what do you adapt the AI tools we'll continue? Yeah. So we can win as an entrepreneur lot of fears dailyap management is biggest subject problem.

1:08:14 subject and when it comes to the employees one thing that we are seeing is what I also want to tell to be able to simplify discussions for many people Third. So first simplifying the language I think will really help in the entire

1:09:02 ecosystem. Oh to simplify all this GPA or you know can I use market can I automate it through 70% at least >> okay >> but building that vision of ours where 70% AI is where you know our brains are working right now. >> Yeah. Questions before the call is ending. Good. >> Good. >> We got the question. Perfect. Yeah.

1:09:50 Okay. Okay. Okay. Right. Thank you. See you. Bye. >> Okay. So based on my understanding the question >> how can we automate >> running ads as much as possible. >> Yeah. >> Okay. So I think we are already in a position where you can automate 90%. is already there right and actually automation where you can buy and sell but which is even the creative part can be automated >> so we'll talk about I I can't show you

1:10:35 stuff I can tell you how to go about it and I can tell you the tools if you want to explore you can explore it'll become very marketingcentric Let's say meta ads meta ads or Facebook or Instagram ads there are two aspects to marketing on ads one that is all ad operations. Yes. Adsets, ad campaigns. What are the problems that ad people solve? Actually performance market.

1:11:39 Maybe ad campaign punch off. We are wasting money on that. If people are not marketers here, they will not understand. I'm trying to simplify as much as possible. But if you're not a marketer, just forward this. >> Uh but that is what you do. >> So both the things can be automated. >> Okay. The first thing, one of the tools, two tools that people can use for automating ad management. >> Mhm. >> Okay. social Facebook ads manager basically which is called as optimization which is look at this scale acquisition budgets on high performance asset. This is what we say right manager

1:12:27 AI automatically detects it and increases it automatic it will it can reduce also. So there are bunch of workflows it can do. Look at this. It's called optimization. The ad optimization >> every optimize reduce targeting purchases. Basically whatever a performance marketer does, >> it automates all of that for you. >> Nice. >> Right. It's actually quite quite powerful. By the way, this tool does one more thing. And the other thing that it does is compet one more thing a performance marketer does is he doesn't only she doesn't only check their own ads but they also check

1:13:13 their competitors ads Facebook ads manager if you go to ad library if ad is running for more than x number of case it >> will someone run an ad which is working which is not working no >> but the duration of the ad is visible on the ad correlation >> not always causation first of all right so it can pull ads from there also right now it can pretty much quite frankly there was a point of

1:14:00 time. This is preAI. PreAI there was a point of time we used to have if you ask DIP we had 15 performance marketers >> now the business is almost 20 times or 30 times larger we have five >> because of ad automation >> ad rules creatives. >> This is your physical product. You want to create ads for this physical product with different people with different backgrounds, people giving reviews of this product. I I use this earphones. They're very, very good. They're amazing. They're awesome. Blah, blah, blah. All of that. Let's say customer review. Amazon

1:14:46 learning positive review. But it's a review on Amazon that will not work as an ad. >> Correct. I can take that same review that is written by a customer. >> Let's say it's his name is Dip and I can convert that into an ad not with Dip's face but with a guy's face. The magic uh >> magics magics. So is that the same tool that you use right now? >> Cool. Yeah, it's the same tool that we use. I'll tell you why a lot of people don't use this >> because they charge percentage of your ad spends. >> Oh, >> they charge 2 3% of their ad spends. in the double catch test now and but a human cannot do what an AI can do. >> I have realized this two years back. >> Absolutely. >> How much are your ad budgets approximately?

1:15:31 >> It's in creral we can also connect our you know ad manager that is also a bas that is what a human marketer does. The action is still taken by a human. Got it. >> When said nanjan said I want to automate it. >> Got it. Got it. >> So you trust an AI to do its job. >> Got it. >> So automated. Good thing about what Nan said is it didn't say 100% automation. >> He said 70 90 >> that 10% still someone has to look at it. >> Correct. >> Someone needs to have a human instinct. >> AI can only understand basis of what it is seeing. There are more things that we see. Right? >> That is where you need that human even now. >> Okay. Performance marketer. Okay.

1:16:18 marketer if they want to win in this AI first world they don't need to know how to change the things they need to know how to ask the right questions >> insight >> nice >> in fact the tweet was about how marketers are the new software engineers >> last generation learning software engineers were the ones who were in highest demand which is still True by the way. It is still true. That is not going away. But if someone can overthrow the demand for a good software engineer, that is going to be people who are good marketers or sales people. When we are living in a world where anybody can build great products, the

1:17:06 most difficult thing, the most difficult thing is to put the product in the right people's hand. >> That can be done by a good marketer and a good salesperson. Someone who understands distribution. So I think it's a platinum era >> for marketing >> modern gold platinum era >> for good marketing sales and good business folks. So now it is basically is the main company which is listed also and first it was like 10 15 minutes. So after that we got to know he's very much into technology. First

1:18:01 technology technology first technology first so I will show you the reason we are not a business we are a technology company selling that's what he believes in sir over to you now so problem statements discuss solutions is there with us >> yeah first thing Hello, I have to appreciate you. Thank you so so much. Amazing. Doing a great doing so great. Keep doing great things. And people I

1:18:49 mean again you know excellent job. Thank you sir. >> Trying to guide for young entrepreneurs founders like you know this is what they all need I'm little bit into technology you know for my business on a day-to-day basis WhatsApp Slack SMS and emails. So the these are the four or five different mediums of communication with the entire team across the length and breadth we have thousands of people between our employees as well as our viewers because everybody is more into technology

1:19:35 everybody knows everything kind of a thing. So I always see I will be missing out on something >> correct >> which would be many things are critical and so many other things are okay to meet. So, but that is again that is not 100% professional. So, I want to be empowered next day. These are the things that you missed out or these are the things that you need to

1:20:20 1 2 3 4 priority will I think youth everybody's flooded with so many things and opportunity will not knock again for anyone it's only when it comes we have to be prepared when it knocks the door absolutely sir So got the question my life is a mess because of so many channels of communication organiz >> okay so let's start with the first one which is let's say meetings >> online meetings online meetings he could use a simple tool called as granola >> granola >> granola or fireflies

1:21:07 >> is it similar to readai >> yes it's similar to readai or aai >> then granola advantage for people who do a lot of WhatsApp calls. >> Mhm. >> Oh, >> so if WhatsApp calls computer desktop app WhatsApp, Granola can take meeting notes of that also. >> Oh, crazy advantage. Okay, that is one. Second for physical meetings or plot the physical device start quick shout out good guys good kids trying to build something called as Neo sapion >> oh yeah so 100%

1:21:52 >> Neoapen okayy it's a physical hardwareh okay >> correct version. They just released Samsung, veryational brand, but I've tested this. It's pretty good as well, which is called Neo Sapion. >> I wanted one anyways. Yeah. >> Yeah. Neo Sapion is a physical device, so people know you're getting it recorded, which is important, >> which is very important. End of the day, it'll give you a summary of everything. And questions, last 30 days, know a machipena aa that is neo sapion for physical meetings right next

1:22:37 you're using gmail slack and stuff now the big question how will I bring all this information to a single platform >> so granola granola key it has something called as MCP >> okay >> model context protocol right >> now using MCP and using let's say NAN >> NA 109 tool you can fetch all the information from every single meeting of granola that's happening every single day. Oh, >> sameap API while API you can pull all the information into a single platform using uh something like N again. Next is this is technical. This is called Google Workspace CLI,

1:23:22 right? Uh it's a GitHub repo. A workspace CLI. Okay. Eng Gmail, me calendar, anything of Google services. Okay. It can pull information from. >> Nice. >> This is workspace CLI very very uh easily. It's very simple actually. It's quite simple information Google services >> anywhere you want anywhere you want. So we got it. We got Gmail covered. We got uh calendar I mean all your meetings covered physical or online. Next is Slack. >> Slack Slack >> which summarizes everything that happened on Slack.

1:24:08 >> Now coming to Gmail. for Gmail also basically information you can use claude claude has something called as MCP so add you can go to connectors directly and add connectors >> for example Gmail you just search for Gmail it's there you just add this and connect your Gmail >> in the same way Slack connectors again add connectors you go to Slack and you can add In the same way, I can also connect granola. Oh. >> Oh, nice. Oh, nice. There you go. Right.

1:25:02 How to how to convert an API into an MCP server. Once that is done, you can connect it here also. >> All you can do is you can come here and just say check my Gmail, check my Slack, check my meeting assistance and tell me what happened in the last 24 hours, right? >> It will give you a summary. Now because we have connectors which is the Slack connector connector every day once this runs use the Slack connector to post this message into my DM. >> So Slack message directly right

1:25:47 >> now the only thing that is not automatable >> WhatsApp >> is WhatsApp. >> Okay. >> And the reason is very simple because WhatsApps are end to end encrypted. M >> so they cannot break that encryption as a result meta doesn't allow in fact meta AI very powerful AI model droper >> right even that meta AI of meta cannot read your messages >> it doesn't have access now does that mean it will not happen I have a way to make that happen also >> okay >> okay it's risky >> but what I would do is I will go to connectors and I will search for Chrome. Mhm. Chrome

1:26:34 24 hours open on the laptop. >> 24 hours. There you go. That's it. She's thinking like a like a orchestrator. >> So, I have my WhatsApp open. >> Yeah. >> Right. >> And it just keeps scrolling and everything. every 24 every 24 hours I will now that has access to my chrome prompt in this Gmail connect but for WhatsApp open my chrome scroll through all the 24-hour messages and give me a summary of that open message once it has access to the browser it will be able to do anything I want to scroll we don't have to go to every individual we can whatever you want. Oh, great.

1:27:26 >> So, what is the risk element here? >> The risk is you're giving your WhatsApp >> last 24 hours. It is not smart. >> So, it will end up opening a message that you don't want to open. Okay. So, connector connect. So I can just say use my browser via Chrome MCP and check the WhatsApp messages in the last 60 minutes and give me a summary. Uh don't reply to any message.

1:28:11 >> Okay. >> Yeah. Let's see. >> Let's see. I've not tried it either. >> By the way, just saying I have done something like this using my open claw, >> using my Jerry. >> It didn't listen to what I said. >> It didn't listen. >> I mean, it didn't listen until I wanted my Jerry to work from WhatsApp via sockets. >> It is supposed to respond only to me. It was replying to me. It was all WhatsApp. Usually I wake up with 50 60

1:28:58 people messaging me in the morning WhatsApp. Hey, do you want to are you available on Wednesday for a talk? say yes I'm available I can this that I'm going to reply WhatsApp group messages everyone who is messaging good morning I'm replying good morning have a great day What is the thing? It made up a budget. You have to pay $25,000 for this. Let me know if you're interested. I'm going to reply WhatsApp

1:29:47 and it's not Jerry's mistake. Then I started to read about it. >> The reason why it's happening is Jerry is not able to understand. It was only supposed to reply to me. Ghani if you still want to use that risky technology today you can I can still give Jerry access to the WhatsApp. Now the idea that I got as we were talking about this is Jerry messages reply that means it can read messages >> which is open claw >> open claw can read messages. So what if I connect my WhatsApp to an open claw instance and strongly recommended not to reply to any messages and your only job is to read the messages.

1:30:34 So every time a message is coming read it and store it. Read it and store it >> and also tag it. >> H you can do whatever you want. Yeah. You can do whatever you want. Yes. Yes. Read it, store it, tag it, whatever you want. And every 3 hours send me a summary >> so it can do it. But don't come and tell me if it ends up replying or something because the technology is too early. So I don't want to even say this in Japan. >> Yeah. >> So that people know it is possible. >> Got it. >> It opened my WhatsApp right now. >> Yeah. >> I said last 60 minutes. Let us see what it is doing. I can't see anything though. It's trying to figure

1:31:19 out the first thing that it's doing is it's trying to figure out what is in the messages >> right how to navigate it. Once it figures out navigate sh second time it becomes fast high level summary. >> Oh nice >> key takeaways looks like family event or puja is happening with photos and congratulatory messages. >> Crazy. So you can do it if you want to. >> Nice. >> The question is do you want to >> do you want to take that risk? No. Okay, this is okay. It's not so much,

1:32:05 but yeah, he's right. Of course, pro. >> No, if you want to use connectors to the full extent, you need max plan $200. >> Mhm. >> Right. But uh logic I mean overall what I tried to do today or what I tried to show you is everything that he's saying is possible using connectors simplest way to do it >> except for WhatsApp. WhatsApp could get a little technical but except for WhatsApp everything is possible. I hope Mr. Prasad finds this answer helpful hopefully. Yeah. >> Yes. >> Yeah. Dr. Hay. >> Hello. How are you? >> I'm great. How are you? >> Absolutely. Absolutely fine. How is your leg? >> All good. Recovering.

1:32:51 But yeah doctor as we discussed uh over to you we want your uh problem statements you wanted us to actually discuss about and maybe solutions is here he's so yeah over to you >> firstly hi Andy my name is Dr. I'm a psychiatrist and psychiatrist one issue that we usually find is 2 days pat. So urgent emergency

1:33:36 usually genuinely what is an emergency is one urgency which they talk about. Correct. for. So we might miss out on something very important

1:34:22 and you know we feel very bad. So is there a possibility they can't say everything to the call center? So is there the way we can use AI where probably they can type in their complaints and AI can prioritize. Okay, this is not that important. >> My second question is I think we spoke about digital addiction, screen addiction, social media addiction and social media addiction,

1:35:08 depressions anxieties. Is there a possibility social media behaviors patterns themes of thinking identify what is the risk these were my two uh issue points that we are facing in today's world and >> got it doctor right so thanks for doing what you're doing I got uh your problems and I think uh I can think of a couple of things that can be done uh which can solve your problem. Uh we're going to explore that and you'll probably see that in the video when it goes out.

1:35:54 >> Thank you doctor. Thank you very much. >> Patient I would uh patsy patients the doctor's voice is the most soothing and the most comforting. So voice hand it off to a human voice AI agent will call the human. So

1:36:41 one call is a emergency serious where the patient is self. So emergency. Okay. What do you think? It's a good starting approach but I feel to solve a problem like this it's not just about getting a reply usually then if we build this chatbot different sent from a chat GPT for 90% of those people who are not serious

1:37:27 >> when non-s serious people first call non-s serious second call non-serious the fifth call of that non-s serious person could be a serious call >> right >> and if he's getting generic advice every single time >> then the person maybe while AI agent the voice agent ability is a very smart thing. I have two things to this voice agent ret. But one problem solve aim personalization. So why do I call a psychiatrist when I

1:38:17 already have AI, >> right? >> Or why do I go to a psych psychiatrist when I already have Google is because I believe that person knows enough about me. >> Correct. >> Right. So answers if it is not personalized to you then it is not very effective is my belief. Second for C for for patient A or should I drink or not drink? The answer could be different and dangerously different based on who you are. >> Correct. >> Based on the history of what? Yes. Correct. What if you are an addict? What if it's okay? I don't know if it is okay, but what if

1:39:03 you can have a drink? So, you'll get a generic answer. Yes, you can drink. Don't worry. But what if you were an addict? >> The answer is no for you. >> Correct. >> Generic bot. Yes. In this case, in case if it's okay to drink after medicine, which I don't know because I'm not a doctor, >> but I'm just giving an example. In this case, context history of the patient is important. >> Correct. So my approach would be while voice agent is a great way to do or we can do a WhatsApp voice agent also or a WhatsApp agent also voice cloning using smallest AI or 11 Labs or almost like the doctor himself >> it will sound like the doctor I have problems with that also >> okay patient will he what will be the perception of the patient when he talks

1:39:49 to the real doctor >> I don't know what it is so if I am the person I would make the AI voice separate and I would make the doctor's voice separate >> right but again we don't know what will work what will not so we have to test it >> paketti if I have to solve this problem I would solve this problem from ground up I will build an app a app where which will maintain all the records of the patient >> in a single place >> doctor's uh and client conf conversations are notes are taken by the psychiatrist >> sometimes is a record also >> for their own note takingaking right >> can I build an app which is completely secure it's like patient and doctor

1:40:35 rules following so the data is not shared with anyone >> where conversations then if I'm a doctor the kind of prescriptions I've given the kind of things that you're supposed to follow everything is documented and created and it's created into an app called using rack Rag rag is basically an AI with a lot of documents access to it. So answer it will look up all the information of you understand who you are understand the latest conversation then we'll give you the answer. >> So for every patient I will build an app for I will build an app which will create a space for every patient >> and I will attach a phone number. So voice agent key call phone number it

1:41:22 knows that it's from patient webhub because it knows that it's from patient webhub. it understands all my problem and it will give me contextual solutions rather than generic solutions. So this would be my approach. Uh and in fact US medical startup which does summarization of conversation and prescription which is a billion dollar company right now. It is made for doctors prescriptions because of so many patients. So this app does the bookkeeping of that. >> It's a massive business. So what I understand from this is if Dr. Vi is

1:42:08 talking about this, there's probably an opportunity. Why does Dr. Vinci have to build this? One of the viewers who are listening to this >> could build this for thousands or hundreds of thousands of psychiatrists that are out there >> which is secure. They have to do it build it with the focus of security with the focus of privacy >> and the best part is coding you can start without code. >> Yeah. where you can v code the products on replet or whatever and get started. >> Okay. MB get started go talk to two three doctors understand maybe you're sitting on a multi-million dollar business. >> Yeah. >> Interesting. >> So yeah.

1:42:55 Hi. >> Hi. Hi. How are you? >> I'm great. So now this is Mr. Aitesh Kurupu topping. We cracked it. >> Nice. So anyways so over to you as we discussed problem statements. So hi Mami. Hi Babber. Uh see basically construction industry uh the biggest problem what you know we want AI or these sort of things to solve is that we whenever we start our sales we commit and end date to our customer right like let's say you bought a flag today I'll say after four years I'm going to give it give the hand over the flag. Now in this four years we do you know like various amount of activities and you can say man hours of labor right >> almost typically average of our communities are like thousand apartments

1:43:41 and each of this community has like 8 to 9 million man hours so these 8 to 9 million man hours will be arranged in a particular sequence that is what will give you end date >> right so our problem is in this four years you know at some point there's a war like how it's happening now at some point the weather is not good at some point election is there because of all this indiscreies our end date keeps changing and for the customer we can only commit one end date we cannot keep changing various end right can help us in accurately predicting the end date

1:44:31 There's a very interesting problem statement. I think it uh goes into predictive analytics. Uh can we do it? I think it's very much possible. We will explore some ways to do it. Do an MVP on how we can test out something with your own data that you have. Uh because you will have understanding of starters. Once we have that data, I think it's we can build a simple model which can uh you know give you an estimate that hey uh this is the safest date that you should give based on all the parameters type. It's more of predictive analytics is what I imagine. Uh but uh I think today models are powerful enough that you don't need a ML team to be able to build this. Right. So >> hopefully >> we will we will we will explore ways on

1:45:18 how to do this in some form of fashion today. But this is very cool. Yeah. >> Yeah. Because like attached to this we can do various automations in uh like purchases and uh you know what do you say uh material dispatch to site like generally when we order material the person who's order material who's ordering material at the site right the store manager and all he will simply order material based on last month but imagine next month So all of those things automatically like site engineer is working on execution like managing laborers supervision all of the and material available.

1:46:04 Ideally the site engineer should only be thinking on how like what is the quality of my work and are the labor available or not rather you should not worry about drawing. So we've also been doing some experiments around this and building some genetic algorithm models. Amazon used some of these things in their delivery whatever when they do package deliveries. >> So we've been exploring but not many conclusive results as of now but hopefully we go there sometime. >> I think I think it's >> we also getting more data. >> Yeah. What the the the question that I have for this is uh do we have data because India real estate doesn't operate on blockchain. So there is no clear data traces there but I think these datas will be in pockets with

1:46:50 companies like yours right as >> a specific uh I would say market level data lay but each company measures a particular way correct correct >> whoever have been measuring like we've been measuring over the last 5 years now we we have like almost 10 million square ft of data only like at which point what did we do and all but yeah still we'll have to you know trying to make meaning out of it >> thank you thanks a lot we'll discuss Thank you brother. Thank you. >> Thank you. >> This is a big uh I think you have to model this out based on data. Uh maybe in their five last 5 years journey 20 projects 20 30 projects 30 different

1:47:48 not a lot actually there are pockets of regions there will be external conditions there will be internal conditions based on what I imagine external condition you will have elections that's an external condition you will have IPL that's an external condition these are all external conditions which are relevant for whole of hydro >> correct >> that is an easier bucket to play then there are internal conditions problem that is that almost becomes like uncontrollable there are external conditions that we usually don't bake in for example uh uh rain m Right. There is no data point that is flowing in but that data is available on API for the next 30 days, 60 days, 90

1:48:33 days also. Will it train, will it not train? >> Estimated. Yeah. >> Right. On top of that only also a lot of external conditions are uh more like you know uh you know let's say in Hyderabad this this periods gets very very hot >> as a result efficiency of workers go down >> right those are clear patterns based on city. Mhm. >> So if you have external data and then project let's say you have done 100 projects 100 projects what were the delays why did these delay why did those delays happen if that part is documented >> I think it's enough to build something better than a guess >> so that's the first step right

1:49:23 you will say four years M >> so your safety net increases you're not guessing >> right and the rational it can tell you the reason why it said 3 and a half years and not 3 years because it will say look three times last four times these are the big reasons now in the future what I'm imagining is this patch of month is going to be very very hot efficiency then there are going to be rains and then there is this bay of Bengal effect that's going to happen >> which we are which is already out there the you know the climate department has already has spoken about it and you have elections coming up >> and then whatever other factors because it can do a deep research to understand what is the impact of this region and

1:50:09 tie all that data together to tell you these are all possible things that can happen correct soat problem Because past these are the 10 big reasons. So they do something called as weightage score that you can talk more about weighted score. What thing leads to what amount of impacted scoring right conditions

1:51:08 WhatsApp condition using that easy. Okay. If there's a person outside, it's able to say there's a person outside, there's a dog outside, somebody left a thing. How? >> So, it takes a picture. Simple. Very simple.

1:51:53 Imagine the same thing is happening. It's the same thing that's happening. Camera only. It uses OCR to send it send it through one of the AI models and it tells you what it is. >> Computer vision models. >> Yeah. OCR. Crazy. >> Yeah. So this is Delhi Mumbai in Mumbai. Mumbai Mumbai. Oh, but thank you. Thanks a lot. Uh time man over to you.

1:52:40 year on production Monday to Monday transaction Tuesday to Tuesday transaction but now that has become a must visit place in Hyderabad If AI can help us in planning production too many functions too many or maybe in way of any which way of I'm sorry to cut you airport okay yeah So

1:53:25 let us check. So last Monday last year most of the 30 40% of the time this is not because of

1:54:14 H last week analysis interesting. No, I mean correlation of uh inflow of traffic at the airport. >> And if 100 people come, two people visit Milofur. >> Right. >> Right. A correlation analysis. And second source is if you're able to get real-time data of consumption on an airport at a flight >> then that prediction is uh to a level possible. >> Got it. >> But it's a very hard problem. I wish we could solve this problem. That'll be a billion dollar problem to solve. >> Thanks a lot. Customer we'll discuss. Okay. Now thanks for your

1:55:00 time. Right. >> See you soon. Airport. Okay. So gueststimate estimated. Okay. So flight tickets data is available via API

1:55:47 average flight That's a good idea. Okay. So, flight watcher flight price watcher basis. Okay.

1:56:42 It's so fascinating there. We're discussing about the data. Crazy. Nice. As we are talking a real estate example, >> will you drink a tea if it is very hot that day? >> Is there a correlation to heat? >> Is there a correlation to weather? >> Okay. >> What are the other correlations that I can find out? Because this is only one kind of data we're looking at. But I don't think this is the only data that we should look at. >> It's all I would look at weather. I would look at sentiment. I would look at

1:57:29 why people are coming. >> Those factors as well. As much data points as I can bring in which I can track it will have no impact. But uh and like India is sitting on an opportunity of a lifetime at this point of time. >> The organized data in the country is less than 5%. >> Imagine which the government has to do. A lot of it has to get organized. A lot of papers on the data we can really really use it to change a lot of things.

1:58:15 >> Hey guys. Hello. Hi. Hi. Hi. Second one. One second. >> Hello. Hi. Hi. Hi. Can you see me? >> Yeah, of course. you. How are you? >> All well, Thailand is burning up. >> First of all, thank thanks a lot. Shoot, you have accepted it. >> No problem. No problem. But yeah. Okay. >> Thanks a lot. So is here is here. >> Hi is here. So I believe

1:59:05 I think he's the first person in in our show where fascinating things line so production law there's a concept called as line production how how do you reduce or how do you effectively use money of the producers you spoke about using AI for script writing possible But then I should tell you this I learned one small one new insight different perspective content I think no one will will speak like that so open

1:59:53 but yeah his perspectives is different so I thought he's he'll be right for this discussion also over to you so as discussed follow statements discuss. >> Okay. So uh nothing much what's happening is in terms of in terms of writing and post-prouction uh what's happening with video generation models there is uh a lot of different uh video generation models uh uh leu there is and lots of other things and there there are these common places where you can integrate all these video models also correct

2:00:38 it's all about control and I think then it'll disrupt something we are seeing inconsistencies with the location character and they are basically definitely becoming better and better one second yeah they're becoming better and better but having said that uh the generative models I don't know if it is even aligned to that perspective locationist that is a lot of uh that's a question that's being targeted by all these video generation models they are making a group of agents also

2:01:24 doing their own custom pipelines correct

…