Hira | Plurality Network (00:00)
course, we always don't market it that way, because whenever I talk about things in this way, suddenly people start thinking about, maybe I'm trying to start a revolution

Ilan (00:08)
Hey, welcome to prompt and circumstance. I'm Ilan And today we're speaking with Hira Siddiqui of Plurality Network who just launched the number one product in product hunt, AI context flow.

David (00:12)
and I'm David.

Ilan (00:36)
All right, on today's episode, we're speaking with Hira Siddiqui. She launched AI context flow. It's an app that was launched on product hunt. It rocketed to number one. And it's a product that really allows you to port your context between different LLMs. So you don't have to rewrite all of the information about anything that you're doing. It's something that I've used. actually released an episode a few weeks ago where we did a first look at this tool.

And in this conversation, we go all over the place.

David (01:07)
There's something for product people, there's something for startup people, and there's something for AI people.

Ilan (01:11)
Absolutely. you enjoy the conversation.

Ilan (01:14)
This week's episode is brought to you by Querio Have you ever found that your data team is bogged down in ad hoc questions from users or that your product team can't quite get the answers that they're looking for out of the Querio's AI agents, it's on top of your data stack and allows you to ask natural language questions to get the answers that you need immediately.

Try today by going to querio.ai, let them know that David and Ilan sent you for two months free of their Explore product. That's a thousand dollar value.

Ilan (01:42)
it's wonderful to meet you. Thank you so much for joining us.

Ilan (01:45)
Can you tell us a little bit about your

Hira | Plurality Network (01:47)
also have almost a decade of experience in IT, IT companies, big companies, small companies, startups, corporates, Fortune 100 companies, the entire shebang. But I've always kind of worked on the what's upcoming new topics. My first job was essentially in a department that was R &D department. And I think I kind of got this. was like

I always kind of felt pulled towards newer technologies, kind of like what's cutting edge, pleasing edge. I don't mind reading research papers and implementing stuff right out of research papers. So that's, that's an area where I kind of always felt like I could shine. So that's why in the past nine years, I've worked on a lot of interesting technologies, Web3, AI, decentralized identity, self sovereign identity, AI clouds. So

very versatile experience, more or less always on whatever is new, whatever is the hottest thing in the market right now. That's the kind of stuff. Yeah, yeah, I think my first job kind of set the scene for the for whole decade. So I just kind of continue doing it. So that was like, yeah. So currently, for the past couple of years, I've been working in Deutsche Telekom as a solutions architect.

David (02:45)
Very cutting edge.

Ilan (02:46)
you

Hira | Plurality Network (03:03)
And this product is something AI context low plurality network. This is something that we have been bootstrapping for a little while and trying to see like how it goes. Trying to find like what, is the one thing that really sticks,

Ilan (03:16)
curious. So basically, if I understand correctly, what you were working on in Plurality Network before was something about storing your identity in a

zero-knowledge way so that with distributed apps who are not using cookies and whatever to kind of track who you are across the entire internet you can still have a personalized experience. Do I understand that sort of correctly?

Hira | Plurality Network (03:40)
Yeah, so that's like an initial idea. So let me tell you like a little background story. So for the past six, seven years, what I've worked on basically is what you can say the self sovereignty of user data and identity. What that essentially means is that today when we go on the internet, our identity and data is like split and siloed into the dozens gazillions of platforms that we use. An average person has seven different profiles we are on now.

10 different AI agents and the entire premise of when the internet was created, when the internet was born, was that it's supposed to be open and everything needs to be plugged into each other and everything needs to work with each other. But data proved to be such a big moat that somewhere, somehow, somewhere in 2000s, everyone decided that it's a good idea to silo as much user data as possible, try to like...

Ilan (04:22)
Mm-hmm.

Hira | Plurality Network (04:33)
have this personalization by making the user generate their data in one platform and then disallowing them to switch into any other. So this became kind of like a competitive advantage and that's why these data silos became bigger and bigger and bigger and bigger and bigger. And we started seeing a lot of negative impacts, negative effects of it as well. So in 2015, 2016, somewhere around this time, the

The same time when web three as a technology was being developed and people were talking about not only decentralization of money, but also decentralization of data and identity. There was a ⁓ parallel group of researchers that was working on self sovereign identity. So these were like, ⁓ but this was like mostly university led and a lot of European Union led research. This is called self sovereign identity. And this is also about like user needs to own their own data user needs to own their own.

identity and be able to take it from any place to any place. when I joined Deutsche just heard about this research area and me and my team member, who's also the co-founder of the Reality Network, we were the people that actually started this small mini department inside Deutsche Telekom. That, OK, what is this new concept? What is self sovereign identity? How do we ensure that people are able to have freedom on the Internet?

we started thinking about more in terms of the open context layer. Consequently, when we came up, when we kind of created this term, this was early 2023.

And when we used to say people like, okay, we are building an open context here. ⁓ everyone was like, what is context? What does context even mean? now fast forward to 2025 context is a word that everybody just instantly gets. So that's also, ⁓ interesting. So we started about things since we had the background of Web3 and AI is between identity. We first started out the first version of the open context here from that angle. That, okay. How do you, how do you make sure.

that user has their data and they can plug it kind of like a USB switcher. So this we always used to give this USB switch example that how like you can take your USB and switch it anywhere. So why can't you do the same thing with your own data?

at some point in the last year, we figured out that, okay, the problem that we wanted to solve is now a 10 times bigger problem with AI agents. How? you talk about data privacy on the internet or data control on the internet, all the data, what was the data that we were producing before we were writing tweets, we were writing blogs, we were writing LinkedIn posts. These are things that are very curated that we know that we are like pushing out into the public and therefore we are

putting some thought before we are like writing something on the internet. And just by your activity on the internet, there can be some inferences done about you, about your political beliefs, about your, whether you're a family man or whether you're some different kind of, you hold different kinds of beliefs about your gender orientation. So you can kind of like infer a little bit things, but it's not that direct. But what's happening in AI agents? Now everyone is just going to the agents and telling

telling these agents everything about their seven generations and their children and their medical history and their kinks and their private lives and their fights with their boyfriends or husbands or whatever. So it's a whole different whole different ball game. And the original thought of keeping the data in silos remains the same. But now what we're doing is now we are giving very intimate amount of information that before was very difficult to get out of somebody.

And now people are willingly doing that. Now think about if this continues in 10 years, what's going to happen? And therefore this is like this is like my personal mission behind what we are building.

course, we always don't market it that way, because whenever I talk about things in this way, suddenly people start thinking about, maybe I'm trying to start a revolution

I'm somebody that's way too much in the idealistic space. So when we market it, we call it as, OK, it's a productivity tool.

Ilan (08:40)
That's amazing. Was there a moment where it kind of like clicked for you and your co-founder that AI agents or LLMs is where we can really accelerate the adoption of this tool and this kind of concept of owning

your identity and your context across different

Hira | Plurality Network (09:04)
Yeah, I think not that straightforward as an aha moment, but there was this one conversation that kind of stuck in my head and I kept ruminating over it for several weeks and then things started connecting with each other. So we were actually creating this decentralized profile system. We were done with it and then we were thinking, OK, what next? What can what next can we do? And then I had a very interesting conversation with a person that I know who, of course, I'm

not going to tell who this person was, but somebody I know closely. And this person was having consistent problems in their dating life for the past couple of years. And then after being rejected again and again, this person actually created an AI girlfriend. And this was like very shocking for me. And we started having a very

⁓ you know, moralistic, idealistic kind of debate that this doesn't feel right, like why are you doing this? And then he said that, you know, it's good for my mental health. And this like every morning I start the conversation with her and then it feels better and then I can work better. And then after work, can, I actually like have a whole routine where I say good night and tell about all my work day and like who was like mean to me or whatever.

And at that point, judged this person a little bit because it felt to me like emotional stuntness But and that is why probably my first reaction was not very good. But then later on, this was a conversation that stuck in my head and I started like researching more into it. And then I saw that there was this one character platform in AI and people like really created their

AI agents there with their long term memory, put everything about their preferences, everything in there. And then this platform shut down and there were people that were so emotionally touched by this shutting down of this platform that this they had these group memorials of their of their companions because it really felt to them like they lost somebody like like a real person. And this felt dystopian to me. But then they

Ilan (11:06)
Mm-hmm.

Hira | Plurality Network (11:10)
I mean, this is the state of the world. Loneliness is an epidemic and this kind of hit me again that OK, what about like these are all the people that are emotionally vulnerable and then we are we are making platforms that gives this intimate details about them to external platforms and then they can shut down. They can misuse this data. They can start advertisements on this and then. ⁓

When I was a teenager, I saw this black mirror episode. I don't know if you if you guys ever saw these black mirror episode But it feels like all of that is coming true Yeah, it felt like like like all of that is coming true and then it kind of like that was kind of like the aha moment that Okay If this is the trajectory that we are we are going Can't we have like an alternative? Technological solution and that is when we thought that okay

David (11:43)
Very familiar with that series.

Ilan (11:48)
Right.

Hira | Plurality Network (12:02)
The same data portability protocol that we have built, why can't we implement it in the AI space? And that is where the AI context flow came

David (12:11)
That's awesome. What a, what a fantastic origin story that was. Thanks for sharing that with us. so do you find that, in terms of those who, who would, be adopting, AI context flow and it's more so in the knowledge worker space,

Hira | Plurality Network (12:15)
You

Okay, that this was also very interesting for us. They are actually people in marketing agencies. then we actually had some interviews with these people as well. And what was happening was that marketing agencies, how do they work? there's like a core group of four or five people. There's like one graphic designer, one content, content writer, one social media manager, and they're like these four or five people. And then they have 10, 15 different clients.

And they need to produce four to five pieces of content for these 10 different clients on a weekly basis. And therefore these are the people that are, and of course, like with AI, the content production has gone through the roof and nobody's like really manually writing everything.

these marketing agencies, people in these marketing agencies, they're context switching essentially 40 to 50 times per day. Because like if they have 10 clients and for each client they have to do two or three small tasks per day, then that's 40 to 50 switches. And these are the people that are also using multiple tools. So.

Nano banana for images, then Claude for writing, then there are other obscure AI agents that they're using as well. ChartGPD for research or Gemini also for research. So they are also using like 5 plus AI tools on a regular basis. So these are the people that actually saw the most time saving and cost saving from this universal context or one context that you can take everywhere.

Ilan (13:54)
Got it, in this use case, are these users basically like, okay, I have 10 clients who I need to produce content for in a week, so each client is its own memory bucket, and then that way I can port that between,

Hira | Plurality Network (14:06)
Yes.

Yes, sometimes sometimes

it's like each client is a memory bucket and sometimes for each client there are three to four buckets. So one for the social content, one for the web copy, one for the research part. there are like, and, then there was also like one odd ICP. This was actually a fitness trainer that was working with 50 clients. And then he created like these buckets for each different client.

Ilan (14:17)
Hmm.

Hira | Plurality Network (14:35)
where he was like putting in, okay, this is the weight, this is the goal, this is the nutritional things that they need to eat. And then, when then they were like asking about, could there be like a WhatsApp bot connected to this so that when they message me, it automatically takes it from the memory bucket, creates a response and then send it. So we're not building a WhatsApp bot now, but that was also very interesting use case that I thought. That people who are dealing with one-on-one with people.

So let's say there's like one, I don't know, financial analyst that has 20 clients, one tax consultant that has 10 clients, one lawyer who has 50 clients. So they are then making like one bucket for one client as well. We didn't think of it this way when we were like creating AI context law, but when you give it to users, they use it in surprising ways.

Ilan (15:21)
Right.

Maybe soon there'll be an N8n node for AI context flow and then you can connect it to WhatsApp and whatever else you need.

Hira | Plurality Network (15:30)
Yeah.

Yeah, yeah, some people also have asked for this. So there are like a lot of ideas in in the air and we are kind of like figuring out, what do we develop first?

Ilan (15:38)
Yeah.

David (15:44)
The classic product management dilemma, yes.

Ilan (15:44)
which sounds like.

Yeah.

Well, also I think a classic sign of product market fit where your users are pulling you faster than you can build what they need.

Hira | Plurality Network (15:49)
Yeah.

Yeah, yeah, we are definitely in that in that area right now. Like there are so many feature requests that my team is like nearing burnout right now and I'm trying to raise and then like I don't even have time to raise because there's like so much other stuff that needs handling and like we don't know what to do at this point. Yeah, but these are the good problems. These are the good kind of problems. I'm happy with these problems.

Ilan (16:07)
⁓

You

David (16:21)
Yeah, yeah, makes sense. So what was it like to hit top spot on Product Hunt?

Hira | Plurality Network (16:27)
It was surreal. mean, I didn't even know about Product Hunt. I had maybe heard about it a couple of times over the past 10 years, but never really knew about it. But then we developed this and then the question came to mind. Okay, how do we reach a mass amount of users? And then I also went to ChartGPT. How do I market?

Ilan (16:45)
Mm-hmm.

Hira | Plurality Network (16:49)
An extension and then one of the ideas that came up that you can do a product and launch then OK, what is product hunt and then we started looking into it. It was was it was said it because I didn't think that it would be like the number one product of the day, but I think it resonated with a lot of people.

Ilan (17:06)
So one thing that I saw you say online was that you're in this process of building up to the launch on Product Hunt that you actually did a lot of work in other platforms as well to sort of like build community and.

build understanding, like shared understanding with other startups and even did that within Product Hunt. Could you talk a little bit about kind of the process building up this relationship with other people who were kind of in a similar space as you?

Hira | Plurality Network (17:37)
Yeah, exactly. So most of the people that are launching on product hunt, okay, number one spot is nice and all, but what's the real goal behind it? The real goal is to actually get users that use your product and tell you whether it's like good or whether it's shit and you need to do something about it. That's pretty much the only reason why you're putting in the effort, right? then okay, there is a product on community that are people that really test out new products and they really try to like, you know,

see, okay, what's happening there. But ideally, you also need to reach people who are actively looking for a similar kind of solution. So where do you find these people? So you can find them on Reddit, you can find them on LinkedIn, you can or different ICPs exist on different, different platforms, but you need to see where your users are living. So what I did was that I thought that, okay, I'm going to reach out to people that kind of seem like my ICP, and then ask them to use my product and then the

I actually sent messages to hundreds of people, like four weeks before the launch. And then out of those hundreds of people, some people tried out the product, some loved the product. So the people who actually took out the time and were like, they thought that the product has some value to it. Then I asked them, we're also launching on Product Hunt, would you like to support us?

So most of the people that were using the product actually went and wrote really detailed comments when we actually launched. So this was like the comments part is not something that you can fake.

David (19:06)
Yeah. What's so, so what's, what's it been like post launch in terms of getting a feedback, you know, like, ⁓ do you, do you, do you reach out directly to those who are using it? is there like a, ⁓ advisory group that, that you set up among customers?

Hira | Plurality Network (19:12)
Yeah.

Yeah, so what we did was that after the product hunt launch, got over the next seven to 10 days, we got around 700 users. And of course, some part of them got joined because there are a lot of people that like try it out and then they just try it out just like that and then they uninstall it. So there was some churn rate there as well. But the people who actually what we did was that we store one thing in the database that

how many buckets is somebody creating? So for example, like how many buckets and how many like, like kind of like usage around 10 days later, what we did was that we went into and we found out these, this Grafana Prometheus dashboard that we have that which are the users that, which are the top 50 users basically.

And then I personally emailed to the 50 users. We actually ran a small pricing test with them as well that, okay, is this a number that you feel like this is like value is $10 the right value for this is $20 mark the right value for this and which agents should we support next? So all of this is something that I'm trying to, of course, like we have like 1500 users. Now I cannot take the recommendation from 1500 people, but

we just use the recommendation from the top 50 people.

David (20:39)
Yeah, that's awesome. the, the, personal email from, a founder can certainly be, ⁓ like very powerful in these early days. Yeah.

Hira | Plurality Network (20:48)
Yeah,

I think that's the... So we do have a LinkedIn page, we do have a mailing list, we have a lot of those things that people say that you need to do marketing. My background is not from marketing, so I have a few people that work that are helping me out figuring this marketing stuff out. But to be honest, the biggest ⁓ pull or the biggest number of users that come are always when I do something personally.

Like when I message somebody, the conversion rate is very different. When I post on LinkedIn, the conversion rate is like, we will get 10 installs. If we do the same post from the product page, we would get maybe one or two installs. So literally the difference is like this. I'd heard that, okay, founder led marketing is something that builds trust, but I was just kind of doing it intuitively. And now I actually see that the numbers are very, very different.

David (21:40)
Yeah, makes sense. ⁓ I heard something similar where, for startups, sales should be, the founder until you get to 1 million ARR. because up until then is not for sure that there's something that's replicable, that you can just give it to a sales team to do.

Hira | Plurality Network (21:52)
Yeah, I-

I mean, if you're far from there, but yeah, I think so too. I think so too, because like,

Ilan (22:02)
Hahaha.

Hira | Plurality Network (22:05)
it should be the founder that should be doing the outreach and sales. The only problem is that when you're starting out like one year ago, I was not comfortable with this role at all. I'd been a developer all my life. I'd written researches. I'd written like I presented on forums, but this felt, I don't know, very icky for me that, you know, I'm messaging people out and it felt like I'm trying to get their time and that it's not polite.

And I had to work a lot on myself to kind of go through over that. And there was this one event I actually took, not a sales class, but it was kind of like an accelerator. And they asked me like, what is the one thing that you absolutely dread? And I said to that person that asking people for money. And this was something I could not just do. And then he said that you have to do one LinkedIn post and you have to

pre-sell your extension, what you're thinking to build for $1. You have to ask people for $1. And it was like, I took three hours writing that one post and I edited it, removed it, wrote it again, edited it, removed it. And then entering my PayPal or link was like, I really had a mini heart attack. My hands were all sweaty, but when I pressed enter, in 10 hours, 130 people actually sent me.

dollars. So that was the point when it completely flipped for me. That was like one experiment. I'm so thankful for that person who pushed me for this. But after that, like that glass ceiling, that glass ceiling has completely like, you know, broken for me. Now I can message anyone. I don't care.

David (23:27)
That's awesome, wow.

Ilan (23:29)
That's amazing.

You

David (23:44)
No problem. You're a

pro now. Yeah. Yeah, definitely.

Ilan (23:47)
I feel like

that's a really good tip for anyone that anyone could take is that, you know, if you're, you know, it's basically like facing up to your fear. you know, this is an experiment you can run and it forces you to face the fear. Like ask for a dollar.

for your product and see what But also I think it sounds like you were really able to communicate the value of the product in the post because if it doesn't seem valuable to people then they're not gonna pay you the dollar.

Hira | Plurality Network (24:18)
Yeah, but then that's a good test. Then that means that you save one year, two years of your life, not trying to go after something that will not work

Ilan (24:25)
Yep.

David (24:28)
That makes a lot of sense. Yeah. You know, it's, it's, it's that thing where, ⁓ sell it before you build it. Right. And it can be, a strong habit to break for people who are technically minded because your natural instinct is to build. Right. And so that's where comfort is. so let me do the thing that's comfortable first. That's fun also.

Hira | Plurality Network (24:34)
Yes.

Ilan (24:48)
This week's episode is also brought to you by N8N.

Its beautiful UI allows any user to go in and experiment with building AI agents, but also create the types of workflows and evals that are required for production level agents.

Try today by visiting the link in our show notes and you'll get two weeks free.

Ilan (25:07)
I wanted to wrap up a little bit with a question that we ask every guest that we skipped past because the story was so fascinating

what's your hottest take about AI?

Hira | Plurality Network (25:18)
Okay, what's my hottest take about AI? The hottest take that I have is that I do think that AI is going to make a lot of jobs redundant, but I don't think that it's going to be a bad thing because I think that people will just evolve and start working on even bigger problems that we haven't had bandwidth to cater until now. For example, space exploration, ocean exploration.

There was this project that I heard about, like there was this team that is using AI agents to decipher how birds talk to each other. So they're trying to decipher the language of the birds. Like when we will be free of the mundane tasks, be it, don't know, content writing, script generation, image generation, then everyone will have to like level up and we will start picking bigger and

better challenges. That's my take.

Ilan (26:14)
I love that All right, getting to the last question here. So obviously you built AI context flow. You're also talking to all these users who are kind of super users of context switching. And so I'm curious for you, if you had to boil down some, you know, maybe a top two or three tips and tricks.

that people should be using when thinking about providing context to LLMs. What are the most important things to you there?

Hira | Plurality Network (26:43)
Okay, since we are talking to a lot of people, we have realized that it's very hard to strike the balance between what everyone needs. So there are like, sometimes there are completely contradictory requirements coming. So there are, for example, let's say there is a developer who's trying to vibe code and application, and they've written their architectural decisions and their requirements and everything in one document. Now they want this entire document.

Ilan (26:52)
Mm.

Hira | Plurality Network (27:12)
to be injected as context when they move between agents so that they can see or they can make some parts of it from one agent and some parts from another agent. So this is the kind of user that wants their entire 100 % context being moved from agent to agent. Then there's another. The content writer does not want the entire blog or entire 50 blogs that they have written. They want a gist of it. They just want a summary. They just want that, OK, this is like,

Ilan (27:36)
Mm-hmm.

Hira | Plurality Network (27:39)
This is the blog that I worked in and this was the direction that I was working in. Now I need to focus on this area. So they need a gist of it. So there are all these different directions that you can go into and it is impossible to make one generic solution for everything. And I believe that for good context engineering, the user will also need to start providing their own preferences. Okay, this is what I need. Like kind of like switches in the tool.

Like I need full context. No, I just need some reason. I just need this because otherwise it's not really that straightforward. The second thing is that when a lot of people confuse context and memory as well, these are two completely different things, not completely different things, but they're sufficiently different. So context is just about, you know, what's in your current conversation window? What did you add in the prompt and what's the external documents that you're providing?

Memory is a completely different ballgame altogether. Memory is more like how the human memory works. For example, when you talk about, let's say you're writing a prompt, you're talking to an AI agent as if it were your friend, and you talk about, what happened on my wedding day? Now, what we're trying to do is we're trying to trigger a memory, a memory of a specific event.

that happened several years ago. if you're thinking about it just in terms of random context, what it would do is that it will say, okay, wedding day, and it would go and it would like run a rag pipeline over your context and then bring in whatever it find related to wedding day. But essentially, what your intention when you ask that question is, that it gives you details about

What you were feeling that day, what you were wearing that day, who came to your wedding, who, what did you eat on your wedding day? And this, these are now relationships. And when, when a human brain works, this memory works in like, like a kind of like, think of it like a graph. So there's this node and then there are several memories attached to this one core memory. So this is like an important event. And now you're, when you ask this question, your intention is to bring these all interconnected nodes.

and talk about it, but a simple context system cannot do that. But a complex memory system can do that. This is a type of memory called episodic memory. And now this is where things get interesting because the memory system of AI is a hot research area right now. There are several attempts to where Chagibiti is doing it in certain way. Claude is doing it in certain way. There are third party external memory systems, but nobody has really got it 100 % right until now.

Ilan (29:57)
Mm.

Hira | Plurality Network (30:23)
And definitely there's no portability even between the vendors that are trying to do it. But this is like, I am personally very fascinated and this is also a problem that we are trying to crack. So that when you, when you like write on a agent, what happened on my wedding day and it, optimize the prompt. It does not just bring whatever it finds related to wedding or day, but like actual events that happened surrounding that node.

And for this, like there are some new techniques like graph rag and stuff that or knowledge graphs that help with this. But yeah, but it's it's a very nuanced topic.

David (31:01)
Wow. I feel like that could be a whole other episode.

Ilan (31:04)
And of course, the last tip is use AI context flow for ⁓ your day-to-day needs to port your memory and context between different LLMs.

Hira | Plurality Network (31:08)
Yes. Yes.

Yeah, and you will get some value out of it today. But if you stay with us long enough, there's a lot of more exciting features that we are developing behind the scenes and a lot more interesting memory features as well so that you can ask about what happened on your wedding day. And yeah, so just try it out.

David (31:30)
That's awesome.

Awesome. So, well, this was a, a great conversation Hira, Thank you so much for taking time to chat with us. Where can people find you?

Hira | Plurality Network (31:41)
LinkedIn. I'm most active on LinkedIn. You just go write Hira Siddiqui on LinkedIn. pretty sure you will. I'll be in the top few results. I'm pretty searchable.

David (31:51)
All

right, sounds good, and we'll also include a link in the show notes.

Ilan (31:55)
amazing. Thank you again. We really appreciate your time Hira

Ilan (31:58)
With that, thank you everyone for listening. That was Hira Siddiqui from Plurality Network. Go check out AI Context Flow and leave us a like, a review, subscribe to the channel and let us know what you think.

Are there other products that you'd like us to test out? If so, let us know. But with that, we'll see you next week.

David (32:16)
Yep.

See you next week.