100x Smarter AI with Meta-Prompting and Context Engineering | Gaurav Bhattacharya

Within a month, we realized that AI is nowhere close to doing the job of an sdr because there's a lot of rejection and there's a lot of, like, human in the loop. You have to go and be creative. You have to go find angles and. And also there's severe limitations on how many emails can you send, how many LinkedIn messages can you send, and if you have a robot doing it, it kind of goes to spam pretty quickly. So you do need bodies on the floor actually dialing and making calls. So we pivoted.

Today I'm here with Gaurav Bhattacharya, CEO of Jiva AI, a company that builds agentic AI agents. He started coding early from humble beginnings in New Delhi, India, and founded his first startup fresh out of high school. He's raised company 10 million ARR and 3.2 million users on the Forbes 30 under 30. He has a degree in aerospace and computer science engineering. So he's a smart guy based up in San Francisco and he publishes a weekly AI sales ops newsletter and he does a agentic GTM tactical advice live LinkedIn calls. He is an expert in AI. Gaurav, welcome. Appreciate you taking the time to talk today.

Of course. James, thanks for having me. So excited to see you and excited to catch up with you again and hang out here.

I love it. So I first found out about Jeeva by one of my VP of sales friends who pinned me and said, well, oh, my gosh, I just took a demo of a new sales tool and you're the demo. You're in their demo environment. You were the person and so reached out. I got to know your company a little bit, got to know you, and. And I'm so excited to be here today.

Man, that's so exciting. That's so funny when you. When you message me that. So. So I'll. I'll give this viewers a little bit more context. So we basically do a can demo. And one of my favorite things to do is I always say for, find me like, heads of sales CROs, people who have sales backgrounds, like experts in the industry. So that was like my agentic query. And I say LA and SF because I split time between LA and sf and you would be the first guy who would come up and I would show like, hey, let me show you how much info I can find about James. You know, like, just ask the AI, like, go do research on James. What companies has he started? What's his sales background? How many. How did he build teams? What has he done? You know, Just give me everything about him. Find me his work email and cell phone. We, everybody, obviously on my team, copied me. So everyone started doing that. So we're just doing that. And it's so funny. So one random day I wake up and so obviously, you know, I'm doing this like clockwork, so I'm not noticing that it's actually you. Right? Like, I'm just doing it. So you sent me a message on LinkedIn. You were like, hey, you got. You guys have to give me some royalty for it. So I was like, who is this guy? Like, what? What does he want from us? Like, and then that's kind of when I put the dots together. The next time I did the demo, I was like, oh, this is James. He reached out to me. That's funny. But what a small world, man. And glad we got connected and obviously look up to you on everything you've done in your career. Super proud of you. So excited to be here.

I love it. Well, you know, two guys who spent a lot of time in AI. I've been doing AI stuff for probably over a decade, but I'd love to learn on your side, you know, your backstory and was there anything I missed in your intro? Like, tell us about yourself and your journey as a founder, CEO?

Yeah, let me do that, James. I'll do a quick background. You did a great intro, by the way. So I grew up in a very blue collar family in New Delhi in India. Lost my dad to cancer when I was just two years old. Mother is a blue collar parent, she washed utensils door to door, so grew up very poor. Humble beginnings, but I was a nerdy kid. I started coding when I was 10 years old. I actually had an uncle who was in tech sales at Microsoft, and he was the richest guy me and my brother knew. So we would always ask him, like, how do we become wealthy? What do we have to do? When we were very young and he was very entrepreneurial. So he told us something that I'm really proud of and I still agree with. And he said that in this time of, in this day and age of the Internet, you got to be really good at sales, you got to learn how to code and you got to do sales, basically. So that's been like my lifelong journey. I started coding when I was 10, built my first video game at 12, started my first startup at 17. But I obviously as a founder, now a third time founder, I've had to learn sales the hard way. It doesn't come naturally to me. But I've been become a sales expert. I've always been the first cold caller, right. Getting the first be the first SDR, the first closer for my teams, getting the first 10 deals, the first hundred deals. So that is my story. My first startup was a very cool like machine learning program for radiologists and we were doing like the old school pre LLM days. We were doing scans and we were taking the scans and automating it. It usually takes like three days or four to five days to do like one report. We were narrowing it down to do it in less than an hour and it was all automated. So imag like a radiology office could produce like really good 97% completed reports in under under an hour. And they could do like 36 a day basically so they could scale operations. And towards the end of it we were doing very hardcore machine learning work as well. We were doing prediction of cancers using ultrasound machines because basically if you use MRIs and CT scans, they're almost 36,000 times more expensive than just doing it with an ultrasound machine. So very different like medical tech. And now I'm doing sales tech which I love. So I would always be the first Salesperson. We closed 500 hospitals. I think I must have close 400 of those in my, in my journey and my first startup. So for the second company we actually had a different idea and we kind of pivoted into gy but we had built this tool internally, James, that we were using for prospecting just like early users of GPT 3.5. And so when our second idea kind of didn't work out, we just pivoted into G by AI and launched earlier this year. But now it's been growing like crazy. Right. So we have like close to 300 plus customers with this product. We have thousands of users all across the world that are using us. So it's been like I had a success where we sold a company for a few million dollars where everyone made a lot of money, was bootstrap company. Then after that I had a high but I had a pretty bad failure with my last startup or actually it's the same company that we pivoted even though we got a lot of, we just couldn't find product market fit. So now I'm onto like my third idea, third startup now. So that's, that's kind of my journey so far. So it's been a lot of ups and downs, but very similar to like every single founder journey. Right. If you think about it, there is.

Definitely the roller coaster, especially with startup. It's a little exaggerated and something no one talks about is that interim period where after you've had a big win, the amount of pressure you feel to do the second one as good or bigger. I've seen close friends get stuck and kind of paralyzed and not wanting to start something again. They get really down when they don't hit another home run. You know, it's. It's pretty cool to see you keep going. And startups are just a game of survival anyways, man.

That's exactly right. Yeah, that's exactly right. It's like I was on a call with a friend of mine who's like a serial entrepreneur. And I don't know if he's a billionaire already, but he could be very much. He created like Google Ads and he's on the board of Coinbase. Name is Gokul Rajaram for People Want to Know. I call him a friend. I don't know if he. He'll say the same, but he's been an awesome mentor to me. But he says this one thing which I love, which is like startups. Startups is like going from one failure to the other with. With the same enthusiasm. It's like you're just going from failure to failure with a lot of enthusiasm and, and one day you'll find some success. So it's kind of very interesting and very true.

So with that, you know, early days when you were doing the medical stuff, we were calling it machine learning, probably calling it computer vision in what you were trying to identify. And now it's AI, right? And I think early way before that it was AI, you know, a decade ago. But what led you to get so excited about artificial intelligence and using OpenAI and using LLMs. What, what was that like?

Man, these LLMs have been game changing, James. And I'm sure, like there's tons of companies and founders like me who have been in the old school machine learning era. And basically like in the early days, machine learning was divided into two aspects. We used to call it predict learning or reinforced learning. And everybody had said that predictive learning is the way of doing it. Like predictive learning, fine tuning of the models, training it with tons of data, that would be like the way of the future, which is going to be a lot of like, if you remember a lot of 10 years ago, thousands of predictive learning startups, predictive analytics startups. Like, that was like what machine learning was, right? You take a lot of social media and try to analyze like social media insights. I'm sure you've seen like thousands of those companies and they. And Big data was a thing, right? Big data. Nobody ever thought that reinforced learning would be so cool. But hats off to Sam Altman. Sam's brother is an investor in us, Jack Altman. So we work very closely with these guys. I mean what a crazy idea. And LLMs have changed the game. Like this truly is AI, right? This is like you don't need tons of infra, you don't need a lot of learning. It just works out outside of the box with just a few fine tune of like what our company does. And you can almost like do everything with it, right? Just like with least amount of training as possible. So I feel this truly is AI. And the unfortunate part is everybody who started in 2020 and before in ML and AI have to like almost reinvent themselves now because the LLMs have completely changed the game, but also a new paradigm. So what I feel artificial intelligence is, is now is I feel 100% is with these large language models. And I know calling them like generative AI and large language models is not the right way to say it because what these models are, is. And they all come from like predicting the next word. I don't know if you've seen the origin of AI is basically saying that if I have a sentence it's going to be autocomplete, it's going to predict the next word for me. But now obviously you can increase the context and it can go and learn multiple things, but that's kind of what it's doing I think. And that's what the future looks like. Imagine in coding and programming we're really good at this. Autocomplete cursor made it really famous, which is what if you have a block of code, can AI predict the next block of code? But then it said that, okay, if you have a block of code and if you have these 36 files, then based on these 36 files, in the context of these 36 files, how can I predict the next block of code? Now if I have these 500 files about Jiva AI for example, then how can I write an entire new function in the program with this block of code? Now this translates to everything else around us, right? In sales, in marketing, what we can say is like here's 500 ads that I've run on Meta and Google and YouTube before and here's ads that have worked for me, here's ads that have worked for my competitors. Now based on this, create an entire campaign for the next month for me, right? Create meta ads create YouTube ads, create Google, create videos for me. It's not far away for us to do that. Obviously it can do it along some humans right now, but that's what AI is going to be capable of very soon. Same thing goes out for sales. James, I can say that here's my market, here's my icp, here's the folks that I've closed, here's the value prop, here's the messaging, here's the kind of Persona and the titles that resonate, here's the messaging that they resonate with. Here's five ideal customers that I've closed. Now I'm going after a similar market in manufacturing. Go create everything for me. It's going to be able to find the next 5,000 people, find the exact messaging and the hook points that are needed to convince them, create like new messaging for them, create sequences, do LinkedIn, automation, whatever is needed to get to a point of, let's say even a sale. I think AI is very close to that, so obviously not there yet. But I do feel like the same examples with coding is going to apply in marketing and sales and almost like every single thing that we do in.

Our life, a hundred percent. And I think you touched on a great point, you know, being able to figure out where you can apply it. And here's the vision of where AI could do and what it can do. But you know, tactically, you know, at the ground level, how are you as a CEO day to day using AI? What are those things you're using AI for as a CEO? You mentioned perplexity a little bit or you know, to find people. But even in the operations of the business, what's working and what's real for you, it could be, you know, really small.

Yeah, James, it's, it's everything for us. I'll, I'll go very tactical on like day to day things. And the caveat being we're a B2B company. We also have like a prosumer arm, so we focus a lot on PLG. So in terms of like sales, I'll start with sales because sales is our bread and butter and obviously not trying to be self promote, not talking about self promotion at all. What Jiva AI is focused on is finding leads, enriching data, doing personalized outreach at scale. That's kind of what we do. Now let's talk about like sales as a use case. I feel the number one thing that we do, perplexity is extremely good at what it can do is create these simple bios and summaries of everyone we're talking to before calls now, anyone can do this. You don't have to have Jiva, AI or any software for this is mostly free, but if you buy a tiered version, it's like 20 bucks a user. Basically what I do is what I tell my reps is you can just copy your entire schedule and paste it into Perplexity for the day or for the week, and it's going to break out every single thing about the person you're meeting. So James, if I was meeting, Go, go read the last 50 webinars or podcasts you've done. Give me your bio. What's going on with your company? How much money have you guys raised? Who's on the board? It can literally give you an entire synopsis of every single data that you want to know based on your company. And then with the memory that you have or Perplexity might have about your business, it can create like a whole snapshot of anyone you're talking to. And the reason that's very tactical and it's extremely important, James, is you know this being in sales, like people buy from people they trust. And the number one way you can build trust is if you're an expert. If you've already done research about the person you're meeting, if you know any ins and outs about their companies and their business, if you speak their language, if you're an expert in their industry, then you can build trust automatically. Because obviously there's one way of building trust is you know the person for 30 years and you're best friends with them, right? Which doesn't happen in sales that often. Relationship selling is very important. But in this dynamic world of more transactions harder and harder, the number one way you can build trust is just be an expert in that person's industry. Perplexity is game changing for that now. Why ChatGPT cannot do it now. ChatGPT is trying to do a lot of these real time crawling and searches, but with Perplexity, it can actually go and crawl 500 different web sources. It can go and read YouTube videos, it can actually listen to podcasts even and take all that metadata and give you like the key insights and synopsis that you need to know. So that is one way that we're using Perplexity a lot. The other things that we use like in marketing or we don't have a single graphics designer, like there was a time. So I have two marketers right now, James. In my last company, I used to have 13 marketers and I feel like, none of them got anything done. Like, I feel like the amount of stuff these two guys with AI are creating is freaking wild, right? So we use three or four or five different tools in marketing. So the number one thing we use is for video. We use something called VO3, which is like a new Google model. There's also Nano Banana now, which is freaking wild to create these images and graphics. But there's a couple of tools that you may not know about or people may. Sorry, the listeners may not know about. There's 11 labs for recording audio, and it does a phenomenally good job. So actually, like, I took a couple of my really long webinars that I had done like two hours long, and I fed the AI. So I have a Indian accent, which is kind of like a bit unique because I've been in the US for some time, so I have like a really weird mixed accent. It does like 100% replica of my voice. And then there's a tool called Synthesia, which is like a better version of hey Gen. And Synthesia can do video. So I fed like three or four podcast videos in different styles, one with a cell phone. I have a one where I'm sitting down and talking like this. And literally what my team does is create these like founder videos that go out everywhere, right? So they're able to take this content literally within an hour or two hours a day. Take Synthesia, take 11 labs, take like something on VO3 and take like, you know, a couple of editing tools and just create like this UGC videos and clips of our customers or for users of me. And that's kind of what we post out everywhere, basically on Google Ads, on meta ads. So it almost feels like we have a production team, but it's nothing, right? Kind of like Riverside, right? Like, have you seen the quality of stuff that comes out on Riverside is freaking awesome. And so that's like tactical ways that we're using this in sales and marketing for programming, man, Cursor is just phenomenal. I love using Cursor. I love using Lovable. I don't know if you've tried Lovable yet, but Lovable is like this ui, like this UI design tool. All my developers use it. Small development team. I'm pretty sure we're doing like 10x output. And that's kind of like the primary things that we do. Like, it's across sales, marketing. Like, the one thing that we don't use AI in James and I actually like it that way, is customer support. Like, we have this One rep who's basically like. It's like a founder Almost, right. He's 24, seven on intercom, looking at email, actually handcrafting responses. Obviously there's macros and stuff that we do, but I've seen that support is one thing. So I'll give you, I'll give you something really funny. And I saw this like I had this one founder who was a really famous guy, I'm not going to name drop him, but he found he had this like whole post about how he automated 99% of like his customer support and it was like really famous. It was trending on LinkedIn and a few months down the line their company just exploded like, or imploded like it didn't work out. It, you know, so, so something happened there and then another founder had done something similar where they were talking about and then in like a few months later found out that everything that they have really high churn and it's not working out. It's like, no shit, you have a high churn, right? Like you're trying to like automate everything about your support. You're not actually talking or taking customer feedback. So that's one thing that we don't use AI in and we still have human supporting like and actually talking to customers and spending time with them. But everything else we're trying to do 99% automation and just like quality checks here and there. So let me pause there. I feel like I went too deep but let me.

That's great stuff. I think a lot of good me. I'm going to check out Synthesia. It's funny, I did 11 labs as well and what I've been doing is you get kind of stuck with your one voice. And so I've recorded multiple versions of myself. Nice games, excited James, professional. And so then depending on who I'm talking to and what region they're in, I can put different content in there and I think that's interesting. I'm very worried that now everything can be cloned, you know and like there can just be a clone of James living somewhere that anyone can say anything. But you know, I like using all this new technology. So yeah, to piggyback on that. There is so much noise everywhere. Right. What are some of the, some of the failures or stuff that's not working in AI that you've either tried or you're like, that will never work or that's all, you know, maybe smoke mirrors.

Yeah, and I'll come out and talk about this.

James.

When I started my Startup. We were a little bit late in the sales game. You understand this really well. You understand, you know, being in sales, you understand like everything that's in our space. When, when I came to market there used to be these AI SDR vendors. And so that's kind of like our first hypothesis was, man, we're so late in the game we gotta brand ourselves as like an even better AI BDR or AI SDR. So what AI? And for, for listeners, SDRs are Sales Development reps. And these are usually like the tip of the spear. They're cold calling, they're emailing, they're LinkedIn messaging, they're taking inbound calls. They're basically like the folks who are setting up appointments or discovery calls or demos, qualifying opportunities. So a really good like low lying fruit or kind of like a low hanging fruit in AI almost feels like too good to be true. Let's automate this function because nobody wants to be an sdr. Nobody wants to talk to an sdr. It's kind of like a hard job. I started my career cold calling so I know like how hard it can be. And people don't usually want to do it. It's a stepping stone for them. So there was a lot of vendors in the space who were basically like growing like crazy. Some guy, some company had raised 50 million. We were like the earliest non. So like just as an example, there were other companies very similar in the tech space too that were saying we're going to automate software engineers. You don't need entry level software engineers. Very similarly, it was like an entry level, you can automate like an entry level QA engineer. So the early hypothesis in the market was that hey, AI would be able to replace these entry level folks basically. And so that's kind of even repositioned our company. And literally within a month we realized that AI is nowhere close to doing the job of an SDR because there's a lot of rejection and there's a lot of like human in the loop. You have to go and be creative, you have to go find angles and, and also there's severe limitations on how many emails can you send, how many LinkedIn messages can you send? And if you have a robot doing it, it kind of goes to spam pretty quickly. So you do need bodies on the floor actually dialing and making calls. So we pivoted to call ourselves like, hey, we're an agentic player now. We're going to help you be 10x better. And I saw that there was a lot of these Software vendors who were positioning themselves as like a full autopilot solution, they all kind of raised a lot of money, but their products are almost like non existent. You don't hear about them. What you hear is like the Lovables or the Cursor or the Windsurf or like Rezero or all these guys who are basically saying that hey, let's make your current staff x better, 5x better, you know, do more with less. So that's, that's one, one angle I know. That's kind of what I've seen. The wave of AI has is like now it's changing to like make your existing team better. So, so that's interesting. Right. That's kind of very similar to the SaaS positioning if you think about it. SaaS is also a productivity positioning. So I wonder whether it's going to get blended in together. But that's like my observation on what has worked in the, in the AI realm, what hasn't worked in the AI realm.

There is this like, I don't know, all software ends up being one giant software if we just kind of let AI kind of roll and repeat. You know, I had a big failure where was using a workflow tool to automate some outbound and OpenAI started hallucinating and putting reference customers that were not actually reference customers in the email. And that's a huge error.

Right.

And it wasn't something you could really anticipate because you think you've already given the guardrails by giving them here, here's our customer, you know, list or here are our reference customers. But it started imagining him.

Yeah.

And so I think you're spot on because to be a successful BDR now or sdr, you do want AI to augment a lot of those tasks, whether that's CRM hygiene or you know, just doing the research. But you don't want it doing the full role. You want the human in the loop. And so what we've done is we just like list out every single task that's required to go from zero to a meeting booked.

Sure.

Where can we sprinkle AI in to help make that person's day more efficient? And then you're right, you can do a hundred x, you can do 10x pretty quickly.

Yeah, that's exactly right.

Spot on. So was there anything else that you wanted to chat about today that you're seeing that's pretty interesting in the market?

Yeah, yeah, maybe. I'll talk about one thing that I find really interesting. There's a couple of Things maybe I'll talk about and you know, for especially like the technical folks on the call could be really interesting. There's a lot of like, obviously chatter about like prompt engineering and like how, how to make it better, how to like get AI to do better. There's a couple of new concepts that I've been learning and we've been experimenting with James. One is called like meta prompting. Meta prompting is pretty cool, which basically means that let's say James, if I have like a prompt and I say that, okay, learn about James's company now. Create emails based on James's profile. It's meta prompting is basically that the prompt keeps improving based on the feedback I get from the market. So if the emails get responded to, then I can make the prompts better. If they don't, then I can change it differently. So that's super interesting. One way that we do it is we actually create like a digital twin. So when we first start working with a customer, we just, you just get their email and in a very secure way we try to read the last 500 emails to create like a Persona. Like if one of your reps hooks up their email to Jiva, they'll know everything about who are they targeting, what's their icp, what's their tonality, what's their sign off, what kind of words do they use, how do they describe the product, do they write short emails, large, long emails? And then meta prompting just helps it keep it, keep making it better, keep making it better based on what they're editing, what they're sending. So that concept I'm really excited about. The other thing that I've been learning is about context engineering, which is super cool. And I'll give you an example. And you should try this when you get time or anyone should try this. But every time you're asking ChatGPT or Gemini or any LLM for anything, just remind them where you are. So if you tell them that, hey, I'm in Los Angeles today and it's like September 30, 2025, then the output is going to be at least 30 times or 50% better than the output if you don't give it like any context. So just context engineering, just telling them your location, time and date, I don't know why, for some reason has a drastically better impact and on performance. So there's a lot of folks, again, like AI does something and people don't know why it does it, but it does a much better job. So if you just pass that, like, if you just talk to ChatGPT today and say it's September 30th, I'm in Los Angeles. Here's my question now, or whatever. You're, you're, you're going back and forth, maybe prepping for, you know, summarization of your board notes. It's going to do a much freaking better job at it if it, if it didn't know that. So, so that's a couple of things that I learned recently that I thought I'll share here.

I'm going to try that I've been doing lately like a daily or a weekly asking Grok, hey, what's happened in AI this last five days? Because it has some more of the context of the X posts that will bring it in. So sometimes it will pick up. It still hasn't been amazing because it just grabs the biggest headlines.

Sure, sure. I love Frog, though.

It's helpful for not having to follow 500 AI influencers.

Yeah.

All right, I got one more for you, which is no one has a crystal ball. But what exciting things do you think is going to happen in AI in the next six months or next year?

Yeah, I truly believe that obviously AI is going to keep getting better. I think the rate of learning, there's something called, I think it's like powers of scale where they're just putting trillions of dollars in AI. And I think it's going to keep getting better. I feel like towards the end of the year, obviously there's going to be tons of these really cool startups and hopefully GY AI and all these others ones can be good and be meaningful. But I feel the really meaningful stuff, hopefully towards the end of the year could be something around gene editing, could be around cancer discovery, or could be around some quantum physics. Because I think the only thing that AI hasn't done yet hasn't created anything net new. Like hasn't created anything from scratch. So I think that's something that I feel should happen by the end of the year. It may not be as big of a discovery, but it could be something net new is like, hey, we solved this math equation that's never been solved before. We created something net new that only humans have discovered. That I think that would be like a really cool, like some invention with AI that is not like a derived. Right. Invention. It's not something that. Because even a picture, like for example, like Dall? E became really famous. Right. I don't know if you remember when Dall? E came out, it was a. Create an avocado chair. Right? So it would take like these avocado chairs, but then what they found out was that it was drawing inspiration from some avocado chair that someone had actually created. And it was in the database and that's what it's learning from. So everything is derived with AI, right? It already exists, and that's what it's using. It's not creating anything that new. So I think it would be awesome if it's able to create something that new by the end of the year. That would be freaking wild, you know, because then the, you know, the. Then you're.

You're onto something there. I saw a. I think it was a USC professor who used a combination of AI and quantum computing to develop a new paint for airplanes. And I think maybe BMW had sponsored this.

Sure.

But the paint is like ultra durable, rust proof. It's like, you know, 100x better than what they're currently using. Use it. Because she was able to use the compute to try a bunch of different combinations really quickly to come up with the best one. So I thought that was. That was a pretty cool. If I can find the article, I'll shoot it over to you this afternoon.

That is pretty.

Where can people look you up if they want to follow your newsletter, your agency, your LinkedIn lives, all of your stuff?

Yeah, totally. So LinkedIn easy to find on LinkedIn. Kaurav Paracharya. I'm actually one of 13 Gaurav Paracharyas. So that's freaking wild because one time I was at a conference and somebody was like, how do I find you? I was like, just look up my name. It's probably like one or two. And he was like, there's 13 of you. Where do I find you? I was like, I don't know, man. That's so weird. Like, you know, like, like, yeah, there's like more James than Gaurav's. But it's crazy that there's so many Garavada charges. I had no idea, James. But on email, I'm G at J e E V AI, so it's GVA AI. I'm super responsive on email or on LinkedIn, so either.

Either of those work amazing. Well, appreciate it. I picked up some great stuff and I think you shared some good knowledge. So thank you so much for the time.

Thanks for having me, James. Appreciate the opportunity to be here. Thanks again.