AI PM is the Job Opportunity of the Decade (Crash Course)

1. 0:00 – 1:21

   Intro

   1. HFHamza Farooq0:00

      There's a lot of hype on AI, but it's actually the opposite when it comes to AIPM roles. Look at the median total comp that is available for them. The median salary for AIPM is skyrocketing.

   2. AGAakash Gupta0:10

      Hamza Farooq, an expert who has worked with companies like Home Depot, TripAdvisor, and Jack in the Box on their AI, and is gonna give you all the sauce for free.

   3. HFHamza Farooq0:19

      Previously, you didn't need to know the technical details. You didn't need to know what RAG is. You didn't need to know what fine-tuning is. You don't need to know how context engineering works. You have to be a jack of all trades. You need to know exactly what AI can do.

   4. AGAakash Gupta0:32

      If they wanna go from here, what else is there in the six-month roadmap to go from no experience to PM at OpenAI or at Anthropic?

   5. HFHamza Farooq0:39

      And within 30 minutes, we were able to build Lovable, connect it to n8n, and have RAG working right in front of us. In today's episode, we're going to teach you everything from prototyping, RAGs, and agents, all how to become an AIPM.

   6. AGAakash Gupta0:53

      Can somebody without experience become an AIPM? Really quickly, I think a crazy stat is that more than 50% of you listening are not subscribed. If you can subscribe on YouTube, follow on Apple or Spotify podcasts, my commitment to you is that we'll continue to make this content better and better. And now on to today's episode. Hamza, welcome to the podcast.

   7. HFHamza Farooq1:19

      It's a pleasure to be here, Aakash.

2. 1:21 – 4:04

   Is AI Product Management Real or Just Hype?

   1. HFHamza Farooq1:21

      Thank you for having me.

   2. AGAakash Gupta1:21

      So the topic today is AI product management, and where I want to start is AI product management all hype? A lot of people have been telling me this role doesn't even exist. A sp- a lot of product managers have been saying people are talking too much about AIPM. Is it real?

   3. HFHamza Farooq1:34

      Well, I'll tell you one thing. There's a lot of hype on AI, but it's actually the opposite when it comes to AIPM roles. Because who's gonna make them into real products, into real systems so that users can associate with that? And I just want to show you one screen, just one detail to show you the growth that we have seen for the r- for the requirement of AIPMs across the board, and it's just, it's just because these things are a need for today. You need to learn all these things so that you can build products so users can go beyond using ChatGPT. And if there's one more, you know, sort of marker for it, look at the median total comp that is available for them. You know, look-- you look at the-

   4. AGAakash Gupta2:13

      Wow

   5. HFHamza Farooq2:13

      ... the 25th prompt, the 75th. I mean, I really hope that, you know, people land at the 90th percentile, but just the median salary for AIPM is skyrocketing. This is only second to software engineers or even exactly where the software engineers are right now in, in the Bay Area.

   6. AGAakash Gupta2:27

      Why do you think that AIPMs are paid so well?

   7. HFHamza Farooq2:30

      Well, it's basically not just a PM role anymore. That's the, the, that's the re-re-reason behind it. You have to be a jack of all trades. You need to know exactly what AI can do. Previously, you didn't need to know the technical details. You didn't need to know what RAG is. You didn't need to know what fine-tuning is. You don't need to know, hey, this is how context engineering works. Here, what we're seeing is that the AIPM is needed to understand to build the next generation of tools that did not even exist a year ago. For example, Lovable. Lovable is a product that we didn't even imagine about, like, a year ago or, or maybe a little more than a year ago. Who would have thought that you could just build websites just like that? Your AIPM role is paramount to make this available for the next billion users.

   8. AGAakash Gupta3:17

      It's really thinking ahead about what are gonna be the next generation of products. These are gonna be AI-enabled products, therefore, we need PMs who deeply understand AI to build them.

   9. HFHamza Farooq3:26

      Exactly. 'Cause if you don't know what AI can do for you, you are stuck in going all about what GPT is doing, what OpenAI is coming up with. You have to think about, "I'm an e-commerce company. How can I use AI to get what's best for my customers?" That's what you need to know what AI can do for you.

   10. AGAakash Gupta3:44

       This is some fascinating data. OpenAI and Google obviously playing at those 90th percentile that we saw. Oracle playing at that 10th percentile. Overall, AIPM, highly lucrative field.

   11. HFHamza Farooq3:55

       Absolutely. Absolutely. Like, this is the best time, I think, ever, uh, it has been to become an AIPM and actually lead from the front so that you can build products

3. 4:04 – 4:43

   Can You Become an AIPM Without Experience?

   1. HFHamza Farooq4:04

      for your users.

   2. AGAakash Gupta4:04

      Does any of the content around AIPM really matter if you don't have AI experience? Can somebody without experience become an AIPM?

   3. HFHamza Farooq4:13

      Well, that's what people have done in the past couple of years. Like, let's say, if you want to become a doctor, you have to go through school, you have to have learning, you have training. Well, you know, to be honest, OpenAI just dropped GPT to us, you know, in 2023, and we had to all-- we all had to scramble and learn. And that's the biggest thing that, you know, that we tell people is don't live in FOMO. There is a roadmap, there is stuff, there is training that you can achieve, assimilate, and actually become an AIPM in, like, six months or less.

   4. AGAakash Gupta4:42

      Amazing.

4. 4:43 – 9:13

   The 6-Month Roadmap to Become an AIPM

   1. AGAakash Gupta4:43

      So what are the key skills? What's the six-month roadmap to become an AIPM?

   2. HFHamza Farooq4:48

      Well, I think the, the most important thing is to first identify the tools that you sort of need to learn to get started. And, you know, if you s- if you l-l-look at the slides, this slide seems a little daunting, like, "Oh my God, I have to learn so many things." But honestly, it's how you use them, and you don't have to use every single one of them. For example, um, and I'm gonna start using the, the, the whiteboard. So the first thing you need to know is what kind of tools do you need and how do you want to use them? And I'm gonna draw a very basic architecture design for you. The, the most simplest, you know, that I've-- that, uh, should get you started. You need to know how to use an LLM as an API endpoint. So you have GPT available to you, have a, uh, you have ChatGPT available to you as an application. They all have API endpoints, so you need to have an API connection, uh, or be able to understand how to use an API, which is almost literally getting a key, and you're ready to go. From that, you need a backend ecosystem. For example, n8n. n8n is this magical ecosystem which is completely no-code. You don't need to write a, a single line of code, and this, this amazing toolGets you up and running with the backend of what the engine that becomes your AI backend. And then like any other tool you need to build, you h- need to have a front end, or the main landing screen where your users will sort of, uh, communicate with you. So if you go here, you basically, what you want is a front end, and we use something like a Lovable to build a front end, right?

   3. AGAakash Gupta6:23

      Got it.

   4. HFHamza Farooq6:23

      And again, this is a no-code tool.

   5. AGAakash Gupta6:26

      LLM backend, front end, pretty simple architecture.

   6. HFHamza Farooq6:30

      So I wanna take you through an example. You know, uh, this, you ... I've written th-three, three different architecture, uh, uh, components. Now let me show you a simple example. So Akash, I wanna ask, do you use Airbnb?

   7. AGAakash Gupta6:43

      Of course.

   8. HFHamza Farooq6:44

      Right. I am the biggest fan of Airbnb, like I'm their, you know, platinum level. If they had a tier, I'm the platinum tier.

   9. AGAakash Gupta6:52

      [laughs]

   10. HFHamza Farooq6:53

       Now the one thing that I hate about Airbnb is I actually, I cannot search anything, uh, beyond the dates and a city. I literally cannot do anything beyond that. So I built this unofficial version of Airbnb and, you know, I really wanna show you guys an, an example of this. So for instance, I want to stay in a modern apartment in Tokyo near the train station. And, uh, looking into today's, let's just say that I wanna go from 25th of November to 2020, let's just say 29th of November. And that's all. All I need to do is just enter my name and my email, and what this tool will do now is it will connect. So this is the front end, which is built on L- Lovable. This is the backend. This is the backend, which has been built on n8n. It will receive the request from the user, and using the request from the user, it will call something which is called MCP. And before you worry about what is MCP and what are these things, all I want you to worry about is everything over here is built with a no code. Like, y- all you need to do is build things in the simplest form. And once you have this thing ready, this workflow gets activated. And once this workflow is activated for you, it will generate a response. It will run behind the scenes, and then it will generate, uh, uh, the, the listings for you that you can find in, um, in Tokyo. And once it has discovered all of them, it will send you an email saying, "Hey, I have found the, you know, the, um, the Airbnbs that are most relevant to what you are looking for." Right. So imagine we identify a customer need. We said this customer, uh, we as customers would like better search capability, and we are able to get those, that search, uh, that search available to us. And, you know, as part of a magician, the last thing is to show the result. So when I open my email over here, I have the da- it picked up the dates, it picked up locations.

   11. AGAakash Gupta8:53

       That's awesome.

   12. HFHamza Farooq8:53

       It tells you why it's a good location for you. And just to-

   13. AGAakash Gupta8:56

       It's like na- it's like ChatGPT natural language concierge for Airbnb. [laughs]

   14. HFHamza Farooq9:02

       Right. And then if you click on a link, it's actually the, like it's not hallucinated for you. It's actually the right place.

   15. AGAakash Gupta9:08

       It's a live link.

   16. HFHamza Farooq9:09

       Yeah, it's a live link, right? So-

   17. AGAakash Gupta9:11

       Okay. So we've seen the end product.

5. 9:13 – 10:05

   Live Demo: Building AI-Powered Airbnb Search

   1. AGAakash Gupta9:13

      Can you walk us step-by-step how we would build this ourselves?

   2. HFHamza Farooq9:16

      Absolutely. So there are, a- as I said, there are three major component. There's a LLM API, there's a backend where you build n8n, and there is a Lovable front end, which you have to work on, right? So in order to get started, let's just say we wanna start with building the backend API. So, you know, the, the, the best thing over here is that I have, you know, finally gotten to writing comments and everything is available for you. So this is what n8n looks like over here. You have n8n set up. You're like, "How do I know I need all these components?" Well, it's actually pretty easy. Once you start designing the ecosystem, you're like, "Okay, there's a Airbnb. I, I need to build an MCP connector." The m- the job of the MC- CP, uh, connector over here is to find, take your request, and be able to connect to Airbnb app and pull information

6. 10:05 – 11:24

   Ads

   1. HFHamza Farooq10:05

      from it.

   2. AGAakash Gupta10:05

      Awesome. Really quickly, if you've been enjoying this episode so far, Hamza teaches two Maven courses that I highly recommend. The first is the Agent Engineering Bootcamp. This has 4.8 stars with over 78 reviews. It's a seven-week live cohort-based course. People at Google, Walmart, and other places are taking this course and learning a lot. The second course he does is Agentic AI System Design for PMs. If you wanna go to that next layer technically on agents, even further than what we're doing in this episode, you should check out that course. Use code AAKASHxMAVEN for a discount, and use my URL in the description to get that discount automatically applied. Now back to today's episode. Today's episode is brought to you by Amplitude. Replays of mobile user engagement are critical to building better products and experiences, but many session replay tools don't capture the full picture. Some tools take screenshots every second, leading to choppy replays and high storage costs from enormous capture sizes. Others use wireframes, but key moments go missing, creating gaps in your understanding. Neither approach gives you a truly mobile experience. Amplitude does things differently. Their mobile replays capture the full experience, every tap, every scroll, and every gesture, with no lag and no performance hit. It's the most accurate way to understand mobile behavior. See the full story with Amplitude.

7. 11:24 – 20:32

   Building from Scratch with Webhooks & N8N

   1. AGAakash Gupta11:24

      Can we build this from scratch?

   2. HFHamza Farooq11:26

      Well, yes you can, but in order for you to make a lasagna, you should know how to bake something simple.

   3. AGAakash Gupta11:32

      Sounds good.

   4. HFHamza Farooq11:32

      And in order to get us started and not get overwhelmed, let's start with the most basic building steps of n- n8n. All right? The first basic step of, uh, uh, using Lovable and n8n together is a webhook. Webhook is this magical thing that allows you to communicate across different tools, and that's what we're gonna do right now. So for instance, the first thing you, you wanna do is if you go, go to our GitHub, I've created a roadmap on how to get started with n8n. You can start f- the most basic one is start here, which is literally just a couple of notes. But here, what I wanna do is I wanna build something a little, little smarter for us, which is able to connect to Lovable, and it's able to run on n8n at the same time.And the basic version of this is that there is a webhook which is, which is the-- your front end. You have an AI, a webhook which communicates with n8n and is able to respond back and forth in real time for you. So the first thing we'll do is, what you wanna do is if you click on the webhook over here, and this is the best thing. There are so many examples that are already created that we don't have to worry about, "Hey, how do I do everything from scratch?" This is the most basic way to, to get started. There's a JSON file. You copy the, the JSON file, and you come to n8n. You can paste the whole thing, and it's ready to go. It's, it's ready to, uh, sort of build from there to what you have been trying to build. So you come here, paste it. That's your workflow right here. Now, I'm not even gonna start with the webhook so that, you know, I'm gonna keep it as simple as possible. How do you get the minimal thing to work for you? You wanna add something which is called a chat trigger. A trigger is basically the way you want an agent to start talking to you. It's ba- what you call to have your agent converse with your ecosystem. So we bring this here. We connect this to our AI agent, and over here, we just wanna mention that this is connected to a chat trigger. There's a chat model associated with that. There's an Open Router here. There's Deep-- We are using the DeepSeek model. So the best thing about Open Router is that you have access to every single LLM that you can think of. It's a-available for you. So you have DeepSeek over here, and when you click on Open Chat, you say, "Hey, how are you?" And it responds, um, "I'm just a virtual agent. I don't have feelings, but I'm happy to help you with whatever you need." Now, you're like, "Hamza, this is so basic. Why, why is this important to me?" I'll tell you why. We have an AI agent. We've added memory. So I can say, "Hey, can you remember..." Actually, you can say, "Hey, I am Hamza, and I like..."

   5. AGAakash Gupta14:22

      [laughs] Me too.

   6. HFHamza Farooq14:25

      Right? So it says, "All right, I don't care if you remember if you like chocolate ice cream." But what I say is, it's like, "Can you tell me something about me?" All right.

   7. AGAakash Gupta14:39

      Okay. It even expanded a little bit.

   8. HFHamza Farooq14:42

      Yeah. So you see, n8n is so intuitive, and it is so powerful that just using one, two, three, four nodes, it's starting to remember who you are. And from, like if there's an AI agent, you can connect to any LLM that exists, uh, through Open Router. Um, I definitely am wanna tell people to start using Open Router because you need, from one key, using one key of Open Router, you can literally select any LLM that is commercially or freely av-available, right? So one key to rule, rule them all is sort of the, the name of the game over here.

   9. AGAakash Gupta15:19

      What is the rules of thumb or mental things you think about when you're trying to choose an LLM?

   10. HFHamza Farooq15:24

       Uh, number one is that I'm-- You know, you wanna start with the usual suspects. You wanna pick up, um, you know, OpenAI. You wanna pick up Claude. You can pick up DeepSeek. They're all great ones. You don't wanna go into something which you never heard of. So rule of thumb, pick the ones that you're familiar with. They're gonna get you the-- They're gonna perform the best, they're gonna be mo-mostly available, and they're gonna produce the results that you expect, right? So here what we have is something which is contained within. Everything over here is contained within. There is nothing which is going to the outside world. So how do we get this to interact with the outside world? And that's where, you know, things become a little in-interesting. We now introduce a new tool which we call as webhooks. And what webhooks do is they connect you to the outside world. So instead of using chat, you connect using webhook, and you can remove your chat experience. So now what we're doing is that we are adding two new variables or, or two new nodes. We have a webhook over here to listen in for incoming responses, and we have a webhook here which responds to whatever has been re-requested, right? So simple enough, we have a webhook that will listen, a webhook which will, uh, it will pass the information to the AI agent. The AI agent will listen to the conversation that has been sent, and then from there, it will respond to the, via the, the, the webhook. And this is the simplest form. Like, this is as basic as, uh, as it gets in order to, to set it up. So what we do here is, so we can make a little bit of mock data for us that will sort of make sure that, you know, we, uh, we are able to get to the, we have the right information over here. This data is pinned. We wanna unpin it. You wanna set up mock data, and you wanna say... [typing] All right. So what we have done is that we have now created a mock data. In the mock data, what we want to do is we just wanna test outThat are we able to produ- like, are we able to get the results that we hope to get when we try to r-r-run this? And what we'll do is-

   11. AGAakash Gupta18:28

       All right.

   12. HFHamza Farooq18:28

       Yep, we'll, we'll fix this. We wanna define below.

   13. AGAakash Gupta18:31

       What are the tips for writing a system prompt well?

   14. HFHamza Farooq18:34

       It's literally the same tips that you have for writing on ChatGPT, on Claude. All you wanna do is you want it to be concise. You want, you know, keep the replies short, in two to three answers. Uh, you want, if the request is vague, give a genuine answer. These are really good tips. Like, they've already built things over here that you can use. And you have memory over here, and in the memory, what you wanna do is you wanna say, "Hey, I wanna take information," and this is... Let's just remove memory for a second just to test out. What happened just now is that you passed a request through a webhook. You set a mock data on a, on a webhook. Now the webhook has information, has a question which is say, which says, "How many, um, how many, um, uh, moons does Saturn have?" And what the LLM did is that it responded pretty much that we would see in the chat. But the difference over here is that this entire thing can now be connected to a Lovable front end. And I'll sh- I'll sh- I'll show you the m-m-magic in a minute. I just wanna add memory over here. I wanna add simple memory. And the beauty about this thing is you're just connecting the information that is coming from the webhook, and you're just p- you're, you're, you're passing the in-information over here. So what you wanna do is you wanna say, "This is the query." I'm gonna just gonna save the query as a key. Usually, you wanna save the information as a session ID. We can work on that if you have time. But basically, memory is where you're storing the past information, as I showed you before. You're, you're saving memory at this point. How do we export this conversation? How do we export this to, uh, to our users further? Or how do we make sure that... Let's say we wanna save memory at a user level. I wanna say that, hey, um, for every given user, I'm gonna save an information. We'll introduce a user ID so that you can, you can see how that, how that

8. 20:32 – 28:16

   Connecting Lovable Frontend to N8N Backend

   1. HFHamza Farooq20:32

      works. Now what we'll do is that we'll go to our, to L-Lovable, and I'm gonna open Lovable. This is Lovable for us, and I'm gonna write a prompt that says, "Please bill me..." All right. I'm mentioning a few things which I us- which I never mentioned before. Like, I'm sure, Aakash, you've used Lov- Lovable before, but here I'm adding a few, a little bit more in-information over here. I'm giving it information on my webhook that is over, that exists over here.

   2. AGAakash Gupta21:05

      Mm.

   3. HFHamza Farooq21:05

      So I'm saying-

   4. AGAakash Gupta21:06

      You copy the webhook URL.

   5. HFHamza Farooq21:08

      Yes. So I just copied the webhook URL, and I said a post webhook, right? And in the post webhook, this is the, the post web- webhook connected to n8n, uh, post webhook. And the only th- And the only, um, body me-message which will be sent.

   6. AGAakash Gupta21:28

      So please build me a finance agent chatbot which is connected to n8n post webhook URL, and the only body message which will be sent is a user query, and you will receive a response from the webhook.

   7. HFHamza Farooq21:38

      Yep, that's it.

   8. AGAakash Gupta21:39

      Simple.

   9. HFHamza Farooq21:40

      Right. And then we let n8n do its job. Usually, there's a bit of a training that you need to do. Uh, you know, it's always interesting to see how results happen in real time. But what it's doing right now, it's gonna build a finance agent, you know, um, as we, as we know, um, what Lovable does bed- best. But it's... Actually, I've given it the information about the webhook, which I'm very interested to see if it starts calling, if it's able to call the webhook that we have created over here. Right.

   10. AGAakash Gupta22:11

       And is it gonna do this, like, securely? Like, if we deploy this website, is it gonna, like, properly hide this so other people can't?

   11. HFHamza Farooq22:18

       Yeah. So what you can do is that you can add authentication. So if you see on n8n, there is authentication options also. There's basic auth, there's header auth, and there's JWT au-authentication. So you can build in authentication also. So let's say, Aakash, you sign in with your email, and you are like, "I wanna make sure that I stick to-- My, my information is only relevant to me." That's where you're gonna pass the user ID as the authenticate-- Or, you know, the, the au-auth keys which is created from Lovable into your, into your system.

   12. AGAakash Gupta22:50

       Got it.

   13. HFHamza Farooq22:51

       Right? This is like, uh, we right now are just scratching the surface. You know, right now all we're trying to do is how to slice and dice our vegetables.

   14. AGAakash Gupta23:00

       Got it.

   15. HFHamza Farooq23:01

       Right? So coming back to Lovable. So we have a finance agent that is alive. Now, usually, there is always, you know, when, when you start working with it, what you wanna first see is can it, can it connect? So I'm gonna write, "Hey, can you hear me?" And what we wanna do and make sure over here is that we have our n8n active, which means it is now listening for a request to connect from n8- from Lovable.

   16. AGAakash Gupta23:34

       Yep.

   17. HFHamza Farooq23:35

       So we come here and we say, "Hey, can you hear me?" So this is the moment of truth. What happens here is that what it did is that it, it will go, and let's just see what happens, right? It went over here and said...

   18. AGAakash Gupta23:59

       Yep, we see the same message in both places.

   19. HFHamza Farooq24:01

       Exactly. So it's starting to s- starting to hear. Like, let's just, let's just start with, with the fact that it did send us a, a request, and it said, if you see over here, "Hey, can you hear me?" Right? So the b- the thing that we need to focus on is match-The webhook respon-- or the, the request of webhook to how we are seeing it coming to us. And the, and the design of the, um... This is a, this is the only technical part that we need to know, is how is the e- when we have the execution part and we received a webhook, we are trying to see what was the message that was sent to us. So you see there, there is a message over here, right? And if you look at this architecture design, this is the entire header, these are the p-parameters, and this is the query that was sent to us, and this is the message that we... So what we are looking for is the webhook architecture which matches this. So the only difference is that we-- es-essentially, our idea is that we want to match the webhook architecture to the architecture that matches what we have received from our user. So here, if you see, there's a body and there's a message over here. So all I need to do is come here, open this. [keyboard clicking] So I'm gonna try to write body dot message. I'm gonna try this again. It takes a bit of a tinkering.

   20. AGAakash Gupta25:48

       And this is to get the message in the right format.

   21. HFHamza Farooq25:50

       Yes, that's all. That's it. Like, there is also code that you can, that you can use to, to, to, to, to... See? Now it-

   22. AGAakash Gupta26:01

       Sure

   23. HFHamza Farooq26:01

       ... it, it responded, right? So, and I know I'm just building something very basic right now. Um, all I want to show is the capability, you know. The text is not coming in the format that we would like to see right now. But essentially, the idea is we have a completely different ecosystem of Lovable that exists on a different, different path. And I'm just gonna come to the, to this one. Here, what we, what we have done is the webhook sort of enabled us to send a request back to n8n.

   24. AGAakash Gupta26:35

       Yeah.

   25. HFHamza Farooq26:35

       n8n is able to respond back, which is the biggest, you know, to be honest, it's, it's always, you know, like you're putting someone on the spot for, for, for a live demo. But it did work.

   26. AGAakash Gupta26:47

       Yeah.

   27. HFHamza Farooq26:47

       Very basic stuff. It did work, right? And then what you can do is when you-- once you come here on, you know, the, the chat conversation and say, [keyboard clicking] um, "Can you tell me historically which finance model, uh, which..." Finance.

   28. AGAakash Gupta27:09

       Which stocks have performed well. Okay.

   29. HFHamza Farooq27:12

       Right. So right now, it's just using the e- the knowledge of the LLM. It is not connected to any tool. It's not connected to any search. It's not connected to anything right now, right? Now, it's answering just like ChatGPT would answer for you.

   30. AGAakash Gupta27:27

       And in this case, I think we had DeepSky, right, in the background?
9. 28:16 – 36:10

   What is RAG and Why It Matters

   1. HFHamza Farooq28:16

      things over here.

   2. AGAakash Gupta28:16

      So the topic we need to all know about and learn a lot about is RAG. Can we learn about what RAG is and see that in action?

   3. HFHamza Farooq28:22

      Yes, absolutely. So let me start with, you know, the most basic question for you. How many doc- how many documents, unstr-- PDF, presentations, things, how many documents do you think you have on your laptop right now?

   4. AGAakash Gupta28:34

      Thousands. Tens of thousands.

   5. HFHamza Farooq28:36

      Right. Now let's take it to the next level. Let's talk about organization with 10 people.

   6. AGAakash Gupta28:40

      Yeah. Hundred thousand, million, yeah.

   7. HFHamza Farooq28:41

      Right. Now let's take it to 1,000 people, right?

   8. AGAakash Gupta28:44

      Millions.

   9. HFHamza Farooq28:44

      You have millions of documents. Millions and millions of documents. It is estimated that 80% of all your data in an organization is unstructured, which is PDFs, presentations, memos, um, all of that is what's stored over there. How do you retrieve that information? How do I find out what was the latest numbers on our, um, uh, on our conversation with client A? There is no way, unless you do a Control + F, find the relevant name of the document, open the document, read through it, and get to the ri-right answer. What RAG has done for us, you know, in, um, in this, in this day and age, is that RAG can consume all our documents, store them, and you can literally search as if you're searching on the internet or like you're doing Google search. You can literally do a Google search on your own documents, and those documents now are now available for you to use anywhere, right? So, and not just you can ask questions about it. It will not return you the most relevant document. It will also return you a TLDR of, "Hey, based on my understanding of document A, B, and C, this is the latest update that has happened, or this is the latest update that you should know about, about this particular customer." That is the power of retrieval-augmented generation or enterprise knowledge management as we know it, and it is one of the fastest growing, growing industries. Like, agents are great, but the com-- like, there's a great company with the name of Glean. Glean started out as a RAG company, and now they've introduced agents and other things, but that's what their major bread and butter has been to, to make that.Now, in order to make n8n use, uh, RAG, I think it's, it's also, uh, it's als- it's, it's quite possible. Um, what-- So there are two, three ways that, that you can do it. The first way is you can basically connect to Supabase. So there's a Supabase vector store, there's a Supabase to- tools, uh, that you can add, and you will have to add documents to it. Um, and you'll have to design a bunch of things with that.

   10. AGAakash Gupta30:53

       Mm-hmm.

   11. HFHamza Farooq30:54

       Of course, it takes a little bit of time. Like, it just takes step-by-step process to do it.

   12. AGAakash Gupta30:58

       Mm-hmm.

   13. HFHamza Farooq30:58

       However, in order to make something work live, let's just try to make it happen re- in real time. So our company, we have a tool that we call Traversal Pro, and all you need to do in Traversal Pro is upload a document. It does all the work by itself. It will do everything for you so once you have logged in. Again, I'm not trying to sell the product. I'm just saying that we found out that it takes a while to build RAG. Why, why-

   14. AGAakash Gupta31:21

       Oh

   15. HFHamza Farooq31:21

       ... why, why don't we build pro... So think, think about it like this. We have these different, um, toolkits avail- uh, the different projects. So let's say you have knowledge repository. You have knowledge repository one, two, three, four, and what we would like to do is make sure that knowledge repository has only the most relevant documents that we want from a customer. So-

   16. AGAakash Gupta31:44

       Mm-hmm

   17. HFHamza Farooq31:45

       ... it's almost like, it's almost like magic that we have this project management handbook over here, right? Um, and what we can do from here is you can go to the playground, you can go to the, go to the playground, you can ask questions. Can you... Hopefully it should work. Right. So what it did is that it got you to that answer reading the document for you.

   18. AGAakash Gupta32:09

       Got it.

   19. HFHamza Farooq32:10

       And it gives you-

   20. AGAakash Gupta32:11

       Thank you

   21. HFHamza Farooq32:11

       ... grounded responses on where did the answer came from, which part of... Which part, you know. Which part of the chap- which page and all of that. It shows you the chunks that it g- came up with the answer from, right? So you're like, "Okay, Hamza, this is over here. How do I bring it to n8n? How do I go from here to n8n?" It's so easy. You go to an API key. You generate your A- API key, right? You save that API key somewhere. I'm gonna save it temporary, temporarily over here. Now you're like, "Hamza, I don't, I don't like to write code. Why, why, why are you making me write all this code?" You don't have to write any code. Here's the best part. You don't have to write any code at all. You can copy this entire code. Let's just say, you know, th- this is the code.

   22. AGAakash Gupta32:52

       Yeah.

   23. HFHamza Farooq32:52

       And you come to n8, n8n, and you say, "You know what? I wanna connect to a device. I wanna connect to another API endpoint." So you select a HTTP request, and you can just tell the HTTP request that this is the... If you could click on Import cURL-

   24. AGAakash Gupta33:09

       Got it

   25. HFHamza Farooq33:10

       ... you can, you can paste the exact thing over, over there. It will figure out what it needs, what you need to, what it needs to do for you. This is the key. Guess what? It built it, everything for you.

   26. AGAakash Gupta33:27

       Nice. [keyboard clicking]

   27. HFHamza Farooq33:39

       All right. So here we have saved it. And what we can do is we can call it... So here, I'm g- just gonna define what this tool does for me, and that's it. [keyboard clicking] All right. So here, this is the thing. We have added a RAG tool right here. The RAG tool is now connected to another backend API. So n8n is this multifaceted tool, I'm gonna sh- come back to the whiteboard, that can also enable you to connect to multiple API services.

   28. AGAakash Gupta34:25

       Mm-hmm.

   29. HFHamza Farooq34:27

       You're not just using just the capability of n8n's orchestration. It has the ability to connect to multiple APIs outside and call that response for you in real time as you are building your product. All right. So let's just hope this works. Um, I'm gonna try to, um, I'm gonna try to update this and say, you can also say...

   30. AGAakash Gupta34:53

       "You have access to a RAG tool to answer questions about product management."
10. 36:10 – 38:48

    Ads

    1. HFHamza Farooq36:10

       right in front of you.

    2. AGAakash Gupta36:10

       Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001, centralized security workflows, complete questionnaires up to five times faster, and proactively manage vendor risk.Vanta can help you start or scale your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company. Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vanta to manage risk and improve security in real time. For a limited time, my listeners get $1,000 off Vanta at vanta.com/aakash. That's V-A-N-T-A.com/A-A-K-A-S-H for $1,000 off. Today's episode is brought to you by NayaOne. In tech buying, speed is survival. How fast you can get a product in front of customers decides if you will win. If it takes you nine months to buy one piece of tech, you're dead in the water. Right now, financial services are under pressure to get AI live. But in a regulated industry, the roadblocks are real. NayaOne changes that. Their airgapped cloud-agnostic sandbox lets you find, test, and validate new AI tools much faster, from months to weeks, from stuck to shift. If you're ready to accelerate AI adoption, check out NayaOne at nayaone.com/aakash. That's N-A-Y-A-O-N-E.com/A-A-K-A-S-H. Today's episode is brought to you by the experimentation platform Kameleoon. Nine out of 10 companies that see themselves as industry leaders and expect to grow this year say experimentation is critical to their business, but most companies still fail at it. Why? Because most experiments require too much developer involvement. Kameleoon handles experimentation differently. It enables product and growth teams to create and test prototypes in minutes with prompt-based experimentation. You describe what you want, Kameleoon builds a variation of your webpage, lets you target a cohort of users, choose KPIs, and runs the experiment for you. Prompt-based experimentation makes what used to take days of developer time turn into minutes. Try prompt-based experimentation on your own web apps. Visit kameleoon.com/prompt to join the waitlist. That's K-A-M-E-L-E-O-O-N.com/prompt.

11. 38:48 – 43:28

    Context Engineering vs Prompt Engineering

    1. AGAakash Gupta38:48

       So RAG is a many tens of billions of dollar industry. People are also talking a lot about fine-tuning and context engineering. What do AIPMs need to know about those?

    2. HFHamza Farooq38:57

       Awesome. Well, I'm gonna talk about context engineering, all right? And context engineering, for me, is the most important thing that an AIPM should know. Usually, what we do is that, you know, when you're trying to talk to, um, an LLM, we have a system prompt and we have a user prompt, correct? These are the two things that we have available to us, or we basically use them t-to, for our LLM to response, respond to us in the way we would like it to, you know, sort of, uh, sort of, uh, uh, respond. However, let's just say, Aakash, I want to make it personalized to you and personalized to me. So let's say we have a finance agent which is listening to your conversation, and you're like, "Um, you know what? I only wanna invest in S&P 500. I wanna, I wanna do safe trading. I wanna do things, you know, that which have historically, you know, been, um, they have gone up and they are safe bets." So we wanna say, "Aakash wants to go after S&P 500," right? And now, l-let, let's just say I'm another person. I'm like, "I'm very frivolous," uh, and I'm more like, "I'm gonna go into crypto."

    3. AGAakash Gupta40:09

       [laughs]

    4. HFHamza Farooq40:10

       Right? I'm like, "You know what? I don't make any money." [laughs] You know, ever since I, I could Google you, realize how m- less money you make. Uh [laughs]

    5. AGAakash Gupta40:20

       [laughs]

    6. HFHamza Farooq40:21

       Right? So I'm gonna, you know, do or die or, like, all in. So how do we make sure that v- you are getting the right information, like, you, uh, uh, are getting the right information, and I'm getting the right information that is relevant to me? And that's where context engineering plays a role. So here, what we do is that we do, we have a system us- prompt and a user prompt. Then what we can do is we add memory, which are your past interactions and things that the LLM has learned about you and stored. So we call this the long-term memory. This is also a prompt. And then we get information, relevant information from the RAG. So j- you just saw the RAG, right? You just saw that the RAG was able to pull context and be-- you were able to get an answer. That's your RAG connector right there. So when you combine all these things, that's your context engineering.

    7. AGAakash Gupta41:15

       Got it.

    8. HFHamza Farooq41:15

       So prompt engineering is what you tell an LLM. Context engineering is how you design the instructions for your LLM, and that's the b-beauty of having the knowledge of context engineering because it then makes your entire ecosystem dance. You can get personalization, you can get, uh, specific answers to what you're, you're looking for, and it understands each users based on, on that information, right? So context engineering is extremely important, more important than prompt engineering now because you have to combine multiple levels of things at the same time, get to, to that answer in today's world, right? So that's what we have context engineering for. The second thing that you have mentioned is fine-tuning.

    9. AGAakash Gupta41:58

       Yep.

    10. HFHamza Farooq41:58

        So you see, fine-tuning is all about task adaptation. So usually, when you have a LLM, the LLM can produ-- You know, you give it a prompt, and for every prompt, there's a response. However, you wanna say, "Hey, you know what? I just, I don't want this prom-- uh, this LLM to produce general responses. I want the LLM to only produce Python code and the best pr-practice Py-Python code." So that's when you fine-tune an LLM on best, uh, best practices of coding, so it becomes a coding agent or a coding L-LLM, right? So fine-tuning, again, I wanna say isTask adaptation, in which you're basically telling the LLM, "This is how I want you to respond, and this is the context behind it. These are 10,000 examples or 100,000 examples of what greatness looks like over there."

    11. AGAakash Gupta42:48

        Mm. And once you give it all those examples, it can really customize to that specific use case and become that specified LLM.

    12. HFHamza Farooq42:56

        Exact- yeah. Another example is that you, you work, let's say you work for a pharma company, and you want the LLM to remember the vocabulary about your things, about the different, you know, acronyms you have and, uh, different, uh, different, uh, products that you have built. That's where you fine-tune for understand- for vocab- I c- want to say vocabulary and not knowledge.

    13. AGAakash Gupta43:18

        Mm.

    14. HFHamza Farooq43:18

        Vocabulary refers to adding new words to the LLM. For knowledge, what you wanna do is you wanna connect to a RAG.

    15. AGAakash Gupta43:25

        Amazing. So we've given people a preview of the

12. 43:28 – 46:08

    Complete Roadmap: Zero to AIPM at Top Companies

    1. AGAakash Gupta43:28

       knowledge. We've given them the demo. If they wanna go from here, what else is there in the six-month roadmap to go from no experience to PM at OpenAI or Anthropic?

    2. HFHamza Farooq43:37

       So I'm gonna share sort of like a very basic architecture that, you know, I, I tell people. You have to learn by doing things. The best way is to keep building things. This is a sort of a basic roadmap. You, you need to know what LLMs are. Uh, you know, in, in this conversation we looked into how we can build applications. We discussed prompt engineering, then we looked into, you know, some kind of RAGs. Now, you just have to follow this roadmap in order to build better products, and you have to do it o- you know, you have to follow this theme over and over again till you get to a point where you have identified this is a roadmap or you've started to feel that I am making sense. And this is how you do it. You know, one of the most important questions I get from people is, "What should I build? Why do I need to... What-- How do I go?" And I tell them, think of three wave approach. The first wave is you wanna build something that saves you time, efficiency gain, productivity gains. I'm not doing anything else. I'm just saving time to do something. Exact same thing. The second wave is you wanna build better quality, better output. And the third is that you are building something which is completely new. So for instance, you have, let's say, Aakash, you meet a lot of people. You need maybe a summarizer which says, "This is the conversation you had with all the people today, and these are the action steps that you need to do." That's your time, cost, and efficiency. The second part is quality, better output. "Hey, AI agent, can you take my video, slice and dice into different parts, and make a trailer out of it?"

    3. AGAakash Gupta45:09

       Mm.

    4. HFHamza Farooq45:10

       The third part is, "Hey, AI, can you actually build an AI recording of as if this is the prototype of the script? I want you to build a script. I want you to have, uh, you know, uh, AI agents which act like human and do the recording for me, and then publish it on my, uh, on my YouTube channel."

    5. AGAakash Gupta45:29

       Very cool. Yes. I should build that. [chuckles]

    6. HFHamza Farooq45:33

       Yeah. Right. So, and we-- Now it just makes-- So instead of just following the roadmap, oh, I need to run Learn Tools, you saw in within 30 minutes we were able to build Lovable, connect it to n8n, and have RAG working r-right in front of us.

    7. AGAakash Gupta45:46

       Yep.

    8. HFHamza Farooq45:47

       Right? That's what you need to do. You need to keep building things and f- see where your products fit in the business problem that you're trying to solve for your customers. That, does it solve a user problem? Does it solve an organization problem? Does it align with your business model?

    9. AGAakash Gupta46:02

       Three awesome questions to ask yourself as you guys go on this AI PM learning journey.

13. 46:08 – 51:14

    Inside Hamza's Business: Traversal AI & Teaching

    1. AGAakash Gupta46:08

       So, I wanna talk to you a little bit here at the end of the episode about you. What is the business of Hamza? How big is the business of Hamza? You were formerly Google. We talked about poking fun about those are some serious golden handcuffs. What does the pie chart of your revenue look like today?

    2. HFHamza Farooq46:24

       So there are essentially two different things which I'm doing right now. I run a startup with the name of Travelsa- Travelsa.ai. Uh, we experimented with a lot of different tools. Uh, we had, you know, initial traction on them also, and we kept building. But where we have s- sort of found a PMF is we have become this company that you can connect, that we... Sort of our tagline is, "Intelligence that runs your data." So imagine manufacturers, um, across US. You have, uh, for instance, you have P&Gs of the world, you have the Unilever, but then there are small and medium businesses also. We worked with-- So imagine we worked with a manufacturer who builds all the boxes for Amazon.

    3. AGAakash Gupta47:08

       Oh.

    4. HFHamza Farooq47:09

       They're n- they're not, they're nowhere small and medium, but they're just like a mom-and-pop manufacturing that basically is producing cardboard boxes for A-Amazon. They had zero data scientists. They had zero insights on what and how much should they produce. They were reactive, which meant th- that, "Hey, hey, we're gonna get a response from a, a request from a customer. We're gonna scramble, and we're gonna..." So they were always, like, just-in-time building. What we did is that we built an army of agents for them that can process that information and give 20,000 SKUs at a daily level that this is the demand, expected demand for them for the next one week, for the next 14 days, for the next three weeks, right? And it saves money for inventory optimization. It gives them better planning. It gives them saving money on raw material because they, they know what they're gonna demand be, so they can purchase it beforehand. So we call this product as Olive, and we are working with Jack in the Box, we're working with Home Depot for similar use cases.

    5. AGAakash Gupta48:06

       Mm.

    6. HFHamza Farooq48:07

       So that's the company that we have. The second part of my life in ecosystem is, um, teaching across various, you know, um, um, various industries. Maven is a huge part of that, my ecosystem. In fact, one of the major reasons that I was able to quit Google was I started Maven, I started teaching on Maven, and Maven gave me enough money, or, like, I made enough money on Maven to get started with that, right? So, uh, I would, I, I like to say 10, 15% of my entire, uh, year revenueComes from Maven and, you know, teaching different courses apart from Maven. And the remaining comes from, you know, working in my organization, you know, working with, with our customers, uh, subscription services for our products, so and so forth. Um, and that's, that's how we have been building slowly our company and learning about, you know, um, actually becoming a AIP myself.

    7. AGAakash Gupta49:01

       Yes. So you're practicing these skill sets day in, day out. That's so cool. Why keep the Maven course around if you have a successful software startup? It feels like most people would just do one or the other.

    8. HFHamza Farooq49:14

       You see, I'm gonna l- I'm gonna let you into a secret. I teach, yes, because I make the money, but I teach because I grow.

    9. AGAakash Gupta49:20

       Mm.

    10. HFHamza Farooq49:20

        So the... I teach two courses. I teach a foundation course, which is a AIPM grow course, because that's my learning on how I have learned to build empathy towards users and customers. And when the, uh, the crowd that we get over there are folks who really want to build something, but they do not have the tools to build that. So I learn from their ideas. I learn from their, you know, thought process. The second group are, is a developer course, which I teach on how, what we are building in our company, we, we build in open space. So we, our entire product is, is in public. So we, they're almost like an extension. So we give them, we work with them in problems that we are facing and how we solve them. So working with those developers or, you know, those engineering managers, senior, senior engineering managers, they uplift my technical skill set. So, you know, uh, there was once, you know, we were sitting with friends and, uh, you know, somebody asked me that, "Hamza, if money was not a problem, what would you do in your life?" And my wife said, "He will still teach."

    11. AGAakash Gupta50:20

        [laughs]

    12. HFHamza Farooq50:21

        Right? Um, I mean, I also teach at, uh, Stan- uh, Stanford SPD. I also, also teach at U- UCLA. I also have a book. They are all so, you know, rewarding in terms of the experiences that I gain, 'cause otherwise you will not... Like, yes, you will talk to customers, but you will, you will not talk to so many people. Like, I met somebody from Airbnb, and he was one of my students in, in, in the course, and he was like, "I think we should take what you have built for Airbnb to Airbnb and tell them, 'Do this now.'"

    13. AGAakash Gupta50:53

        Love it. Amazing. Hamza, thank you so much for coming. If you guys wanna check out his Maven courses, you can use code AAKASHxMAVEN to get a discount. He has two courses, as he said, Agentic Engineering Bootcamp and Agentic AI System Design for PMs. I recommend them both. Hamza, thank you so much for being on the podcast.

    14. HFHamza Farooq51:12

        Absolutely, Aakash. It's a

14. 51:14 – 51:52

    Outro

    1. HFHamza Farooq51:14

       pleasure.

    2. AGAakash Gupta51:14

       So if you wanna learn more about how to shift to this way of working, check out our full conversation on Apple or Spotify podcasts. And if you want the actual documents that we showed, the tools and frameworks and public links, be sure to check out my newsletter post with all of the details. Finally, thank you so much for watching. It would really mean a lot if you could make sure you are subscribed on YouTube, following on Apple or Spotify podcasts, and leave us a review on those platforms. That really helps grow the podcast and support our work so that we can do bigger and better productions. I'll see you in the next one.

Episode duration: 52:01