AIMinds #028 | Shai Unterslak, Co-Founder at Based Social Company

About this episode

Shai Unterslak is a software engineer and co-founder at Compass, the viral AI wearable that transcribes your life's conversations, and transforms them into actionable insights. Originally from Cape Town, Shai moved to the U.S shortly after exiting his first company to one of Africa's biggest fintech startups.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this AiMinds episode, host Demetrios talks with Shai Unterslak, co-founder of Compass. Shai shares his journey from starting a company in South Africa to selling it to a major fintech firm, and his subsequent work in hardware-software integration in the U.S. He discusses his early experiences with a payment company, which shaped his approach to product development.

Shai then details Compass’s evolution from a hyper-personalized AI project to a device that turns audio interactions into actionable insights using advanced AI. He covers the social implications of introducing such technology, including privacy concerns and its impact on personal and public interactions.

The episode wraps up with a look at Compass's future potential, Demetrios’s personal experience with the device, and broader thoughts on AI's role in enhancing human-computer interaction.

Fun Fact: The device named “Compass” emerged from a pivot from an earlier project called "Everything Bagel," which aimed to aggregate a user's digital footprint but eventually led to the realization that a significant context of a person's life was not being captured, leading them to focus on audio contexts.

Show Notes:

00:00 First experience in physical products, challenging, interesting.

03:29 Founders worked on "Everything Bagel" data project.

07:41 Transcribing audio and structuring for user value.

13:07 Mute button adoption: Early and unexpected interest.

17:04 Excited to test, reach mainstream, solve problems.

19:06 Cautious about adding features, believe in digital identity.

More Quotes from Shai Unterslak:

Transcript:

Demetrios:

Welcome back to another edition of the AI Minds podcast. This is a podcast where we explore the companies of tomorrow built AI. First, I am your host, Demetrios, and this episode is brought to you just like all the other episodes by Deepgram, the number one speech to text and text to speech API on the Internet today trusted by the likes of the top conversational AI leaders, enterprises, and startups like Spotify, Twilio, NASA, and Citibank. We are joined this episode by Shy, the co founder of Compass. How you doing, dude?

Shai Unterslak:

Hey, man. Good to see you.

Demetrios:

So you've got an awesome story. You started off your journey in South Africa, created a company while you were going through the motions of school, right. While you were getting your education, and sold the company real quick to one of the largest fintech firms in South Africa. Tell me about that. How did it feel to sell your first company? And most people, I think, would love that outcome, for sure.

Shai Unterslak:

Yeah, it was. It was really crazy for me. Like, the highlight was the people at the company that acquired us were, like, my tech idols. So, I mean, in South Africa, there aren't many kind of, like, crazy success stories in tech, and so the people that actually, like, do crush it and grow really quickly and, you know, like, you look up to them. And so I got to, like, work with people that I was super impressed by and could learn a lot from. So I was, like, just excited to kind of work with them.

Demetrios:

So you got to continue your education not only in school, but at the workplace.

Shai Unterslak:

Absolutely, yeah.

Demetrios:

Now, I imagine that planted the seed of hardware and software in your mind. You moved from South Africa to the US to continue your founder's journey, right?

Shai Unterslak:

Yeah, absolutely. So it was my first time working on a physical product. So for context, the company that I worked for, that I joined, they are payments companies. They have physical card machines, pairs with an app. And so for the first time, when I was leading product, I had to think about, like, when should we start speaking to the manufacturer? When should we think about this certification and that certification? And so it was like a totally different kind of product, and the software piece became less and less important. And that was really interesting, because how do you keep users happy across hardware and software? That's like a different kind of problem. I really enjoyed that. It was pretty interesting to me.

Demetrios:

So you came to first New York or no, Austin, right?

Shai Unterslak:

Austin, yeah.

Demetrios:

And quickly moved to San Francisco to live out the Silicon Valley dream.

Shai Unterslak:

Exactly. So I was one of the people who fell for the propaganda that Austin is the next San Francisco or the next, Silicon Valley.

Demetrios:

And you got there and quickly decided.

Shai Unterslak:

No, it is an amazing place. It is great. I love it. I would raise a family there, but, yeah, I just didn't feel the same kind of energy that I felt when I visited San Francisco. So I was like, I'm so close. Might as well just come to San.

Demetrios:

Francisco and tell me the story about compass.

Shai Unterslak:

So compass was a bit of a roundabout story because me and my co founder, Zach, we were actually working on a different product. And we were working on a product called the Everything Bagel. And the idea was, what if you connected all of your data sources online, so your whole digital footprint into kind of like this one? We call it your everything bagel, and then you can do cool stuff with it. And so we would index, structure, embed all of your data, and then you could have this hyper personalized AI friend or AI agent. And that's kind of how Zach and I started working together. And what we realized pretty quickly was the data was interesting for the initial personalization, but it turns out you're not posting every single day or every single thought. And so it becomes stale pretty quickly. So we started to look at other data sources.

Shai Unterslak:

How can we make this feel more alive? And around the same time, we were about to do a major pivot. Our roommate, Martinez, totally blew up on Twitter with an AI wearable launch. And so he launched this necklace that I'm wearing right now called Compass, and it blew up. And he sold, I think, like, 150 in the first week, 180 in the first two weeks, just from one tweet. And he reached out to us and was like, guys, you have been building AI consumer apps. I'm building this AI hardware product. Like, let's work together. And we literally lived in the same building, so it wasn't too difficult.

Shai Unterslak:

And we started working together, and it ended up being like a really cool, cool relationship. He's now a co founder as well.

Demetrios:

And it seems like you were a great fit. Also, with your previous experience in the hardware space, did you not have any fear on jumping into hardware? Because, as they say, bits and bytes are much easier to work with than atoms.

Shai Unterslak:

My take on hardware is like, I think if you're inventing something, it's really tough. So if you're trying to create a totally new type of physical thing, that I think is really tough when you're just trying to make good use of existing tech. Hardware actually, to me, is way more intuitive than certain other types of products. It's like, it's just a thing you have to get it manufactured, you got to make sure the timing works, you got to make sure cash flow works. We're not inventing. This is just a bluetooth microphone. And so some people try frame it as this innovative AI wearable. But, like, our magic is actually, like, what we do with the data source, so we just view it as a data source.

Shai Unterslak:

It happens to require some hardware. And luckily, Martinez is actually, like, a cracked hardware guy. So he is the third co founder. He is probably up at night thinking about hardware more than I am.

Demetrios:

Well, this makes sense, shifting from everything bagel, where you recognized data is valuable, people want these insights, but after you ingest the initial data. So I am assuming you would take all of my posts, you would take all my messages, you would take all my notion documents, all my emails, and you could give me some kind of a story, but that doesn't tell the story of my life. That almost, like, just tells my Internet story. And there's a lot more dimensionality to me as a human. And so the idea with compass was saying, how can we get out of the digital footprint or the digital history of someone and get into their real life?

Shai Unterslak:

Exactly. Yeah. It's all about context. And it turns out a ton of your context is actually not recorded anywhere. And to me, that's actually kind of sad. It's like there's a whole portion of our lives that is just undocumented, and we're getting way better at being able to make sense of unstructured data and tons and tons of data. So it felt like the right time to kind of just make sense of all of this audio data that just gets lost every day.

Demetrios:

Now, tell me how you are thinking about the gigantic amount of audio data that we have and siphoning out or filtering the gems.

Shai Unterslak:

So, it all starts with the transcription, right? Which I think you know a bit about. But we, the first step is we just take all of the audio that we hear, and we try to figure out, first of all, if someone's speaking, if someone is speaking, we will transcribe it. And we take all of those raw transcriptions, we store them, we embed them, and we start to structure them. And the structuring is where a lot of the actual user facing value happens. And luckily, large language models are pretty good at structuring them. And so then it just becomes a product problem. Like, what kind of structure do you want? So, for example, do you want an hour of conversations turned into a three bullet summary, or do you want it to be an essay, or do you want it to be a couple tweets. And that's where we've kind of gotten creative on product, is providing the base level expectations.

Shai Unterslak:

Like, you probably want summaries, you probably want action items, but then maybe you want something else. Maybe you want negative statements to be reframed as positive statements. Maybe you want to understand if, maybe you want to fact check it. And so that's where we've started experimenting with things like workflows that give users the ability to build their own kind of prompts and choose from prompt templates and stuff. But that's really the big unlock is like taking LLMs and taking all of these transcriptions and then providing interesting, like, product experiences.

Demetrios:

And inside of the product, how are you thinking about how to present these conversations or these transcriptions to people? Because I can imagine it can be seen as like one continuous day, one continuous week. Or maybe it's you were talking for 30 minutes on a call or this podcast, and then you stopped and you were silent for a little bit, and then you had a conversation with your neighbor, et cetera, et cetera, for sure.

Shai Unterslak:

So, I mean, there's still a lot of work to do on this. And our goal would be to try mimic how people retrieve things when they try to remember it by themselves. So when you try to remember a conversation you had with a friend, maybe you think about the location, right? Maybe you're like, oh, I remember we were at, you know, Ratchet for there. And you start with that, and then you start picturing the surroundings, then you start to remember. So location is a big piece. You can actually see where the memory took place. And then the other pieces we actually just take. We take a bunch of conversational data, we separate it into separate conversations.

Shai Unterslak:

So a conversation could be an hour long meeting, it could be a ten minute quick chat. But as soon as the context changes, you actually find a different conversation will appear in your UI, and you can tap in and read a summary and stand out quotes and action items.

Demetrios:

And the way that you know if a conversation is changing is because the people that you're speaking to changes.

Shai Unterslak:

It's a combination of a couple of things. So location is one. So it might be a signal, you know, if you move from one location to another, maybe you're entering a different context. But the major thing is just using the, the actual substance of the conversation to decide kind of intelligently. And this is why large language models is such a critical piece of. This is like, I wouldn't know how to write an algorithm to detect when a conversation changes, but luckily large language models are pretty good at that. That's not the end of it. There's still, how do you group conversations? Because maybe I have 20 conversations at a conference, but I want them all grouped so there's kind of more to it than just throwing it into an LLM, but like, yeah, they're pretty good out of the box at that.

Demetrios:

Speaking of conferences, I think that's a perfect segue into my experience with compass, and I was using it as a mc for the AI engineering world fair that recently happened, and it was awesome to have it. I think a lot of people also that were tuning in online were able to quickly skim and get an understanding of what was happening on all the tracks simultaneously. So that is huge. What I realized though, when I was wearing it is I'm a bit of a liability while wearing something like this, and especially if it's being broadcast out to people that do not know me or more than my like, circle of friends. Have you noticed that people are using this in different ways and having a social awkwardness, much like the Google Glass when it first came out? Has there been talk of that? What have you been seen?

Shai Unterslak:

For sure. So, first of all, I just want to kind of mention that the conference piece is very abnormal and was very non compass like thing to do. It was just a growth hack, pretty much. Let's document this conference that, summarize it and provide some value and get in front of some potential users. But the whole idea of compass is that it's your own private repository. So as a compass user, all your conversations would be in your own private app. You can delete, you can edit transcripts yourself. So that's kind of an important piece because I sometimes say stuff and I'm like, I don't know if, even though it's just my compass, I don't know if I want that documented at all. And then obviously we've built features to support mute.

Shai Unterslak:

We have a mute button in the app, we have a mute button on the device, that kind of stuff. But to answer your question, because I think it is a really interesting question, and that's something I was concerned about, is it going to change how people act around me or what I say or what other people say? And it turns out that you'll get a couple of jokes. People will be like, is that like, especially in SF, because everyone has seen these and knows what they look like, and they're like, is that like going straight to the FBI? So you get a couple jokes, but then it turns out people forget and they move on, and you find yourself having, like, your usual organic conversations. It's definitely, you know, at the, it's early on in terms of, like, the adoption curve. We don't know when it's gonna be, like, expected to be wearing one of these. It's still a little bit weird, for sure. And so a lot of our early users are like, early adopter types, but we are seeing some adoption in random communities that we would not expect. Not the typical tech AI early adopters, but, yeah, it's pretty interesting.

Demetrios:

Like what?

Shai Unterslak:

Like 50 to 65 year olds in the midwest? Wow. It's pretty interesting.

Demetrios:

Wow. So one part that I love about this, and I felt like Washington, so useful once you have it, is there are many times that I remember things, and I'm not sure how well I'm remembering it. I, of course, am very confident with my memories. My wife, however, has other opinions on my memories. And so something like compass is great because we can go back and look at transcripts and say, see, we said this, or see, she usually can tell me, see, I told you that we had this thing tomorrow and it's here and do this. And so then I don't have to pester her. I can just search through the transcripts. And so there's nothing.

Demetrios:

And by the way, there's nothing that will get you really in the good graces of your wife, but more than telling them they are complaining when you're trying to fast on a complaint day, a complaint free date. But that's all nother story. I would love to outsource that to some tool so that I can be notified when I'm doing it and then recognize, okay, I'm judging or I'm complaining. I'm doing this. I said I wanted to work better or I wanted to hold myself to a higher standard. And now I have something that is very non biased, non judgmental, just helping nudge me along and keep me centered.

Shai Unterslak:

Absolutely. Yeah, that's it. Sounds like we should send you a compass.

Demetrios:

I already have one, man. Oh, yeah. I got it from the conference. And I've been afraid. As I said, I felt like I was a little bit more of a liability having it, because I didn't know if it was continuously broadcasting things out or what. So you telling me that it's my personal compass, nobody is looking at it. That is reassuring, for sure.

Shai Unterslak:

Yeah. I think starting the user experience with compass at the conference and then going to personal use, I can imagine being a super weird experience. But yeah, I can assure you that I've removed your conference host role on the database.

Demetrios:

So all of a sudden, I'm not getting conference postings. Now, three weeks after the conference, and people are wondering what's going on here. So what are you excited about for what comes around the bend? What's keeping you up at night these days?

Shai Unterslak:

Yeah, so we're really excited to test with other audiences. So I mentioned a couple, like, surprising demographics. I think the only way for this to be a mainstream product is for it to make sense. And people outside of our AI tech bubble, I think we're all curious about what you can do with this data, and that's why a lot of people will buy these products even though they don't actually have a good use case. So finding and really cementing our place in the market with an audience that is actually trying to solve problems that aren't just AI is cool and we are finding that, but just going deeper with those people, we now are growing pretty quickly and just having more of these conversations and figuring out how we can help people and provide real value, that's what I'm very excited about. And also ramp up production. We still have a couple of people waiting. We decided not to be like the other people in the space and have these huge wait lists, and we actually shipped units.

Shai Unterslak:

We made them, we sent them to users. But there is still a process to go from making a few hundred to making tens of thousands. So I'm excited to get to that point where we are rich in stock and we can kind of grow as quickly as we want to.

Demetrios:

Well, it feels like there are a lot of times that you're going to miss out on context because there is no talking. Like, if I think about my day, I'm writing in slack, I'm writing emails, I am doing many things that all you're going to pick up on the compass are keys typing away at the keyboard, or I'm reading a book, maybe I'm watching a movie with my headphones in. Right. And are you thinking about how to capture that data also because it could inform or give more context to compass, and especially since with everything bagel, it's almost like you were doing that already.

Shai Unterslak:

Yeah, absolutely. We're being cautious of adding features too quickly, but yeah, I mean, we still absolutely believe in our original thesis that your digital footprint and all the things you do online is a big part of who you are. It's not the whole part, it's not the most lively and most dynamic part of you, but it is a big part of you. So, I mean, we definitely, we think that's important. We still feel like we haven't scratched the surface and how much value you can provide just from audio. So, you know, it's probably going to be a minute before we start exploring other data sources again, but we're, like, well positioned to add them back.

Demetrios:

Well, dude, this is exciting. I'm going to go dig up my compass and start using it because I want to use these workflows once they come out and keep myself on track. Remember to do things. Hopefully give myself reminders to not complain or judge or just be a better person. Be nicer. Hopefully it can make sure to notice when I start yelling. The other day I was getting real triggered on a call and it would have been nice to have something just say like, hey, take a few breaths. Am leaving that up to my own regulation skills right now.

Demetrios:

And despite trying very hard, I still slip up. So it would be great to set some guardrails and I think that I want to at least experiment with compass and see if it can help me out.

Shai Unterslak:

Awesome, man. Well, yeah. Let me know if you have questions. I'm excited for you to try it again.

Demetrios:

Excellent. Great talking to you, Shy. Sweet.

Shai Unterslak:

Thanks for having me on. Good to chat.