Reclaiming Time: A Podcast AI Approach

When it comes to podcasting, many people feel like they're drowning in a sea of audio. With so many options available, from bare-bones recordings to high-production-value shows, it can be overwhelming to know where to start. But what if I told you that with the right tools and approach, anyone can create a high-quality podcast without breaking the bank? That's exactly what our guest is all about – building a platform that helps people reclaim their time and turn them into 10x producers.

My friend, you're really excited for what we're building yeah 100% basically reclaim so uh again J Jason had tweeted something like I'm reclaiming my time I'm going from Seven podcasts a week down to four um and that that kind of when we were building the company that sort of became our slogan like let's let people reclaim their time and turn them into 10x producers so yeah you know that really is that really is the thing. And with podcasting, it's like there are a lot of steps but the truth is podcasting can be as Bare Bones or as high production value as you want so for someone who really wants Bare Bones you know record zoom and flip it to YouTube that's it right you don't even have to describe anything you just put a title you're you're done but that's you know that's not going to do well what what you want is the production quality you want chapters you want summaries descriptions you want all of that good stuff you want a nice show website that's completely automatic you want to get the word out yeah exactly so you really have to you know up the game and and that's what we've kind of been seeing with podcasting right everybody's upping their game and the idea is for at least for the least work possible make it so someone can up their game with podcast AI you know up to that top level well man I appreciate this

I want to leave with one final question because you are building in the space and AI is so new you're doing all different kinds of things with AI whether it is transcriptions it's or it's so you're kind of doing the whole pipeline I would say not just when you look at the podcast pipeline but when you look at there's audio coming in it's being transcribed you're taking insights from that transcription and then you also have the ability to create audio on the other side of it so you can go text to speech and so for me that that's basically the whole Gambit now what are some of the main challenges that you have faced while building with all of these different new technologies

so the technology itself is unbelievably simple to use I'll just give you a few friction points that might have happened that did happen so one was you need a vector database and you know because you need a vector database um we had to switch from MySQL to postgressql um so that was a little database change that happened very early on because you get PG Vector which lets you um do those Vector searches so that that was one thing um the second uh big issue was with open AI getting access to rates like tokens per minute and stuff like that that was super important for us getting access to GPT 4 really early on was super important to to us and it's sort of hard to wiggle yourself into into that thankfully we recently got a really big uh limit increase but yeah that was that was a big pain point um but it's interesting how the actual working with the llms is just such a pleasure and and probably the literally the easiest one is deep gram because nice you know for podcast AI everything starts with the transcript right so you're uploading that episode and then you're hitting transcribe or it's automatically doing it for you and that's the basis on everything because every other lml is fed from that transcript so the fact that deep gram does the speaker identification identification that's like absolutely critical um and that lets us rapidly identify the speakers the hosts and the guests um and then everything else is from that the the chapters are done from the transcript the summary is done you know from from the chapters um so that's like a little a little cheat to to to go faster um and and and so so really the root of it is is the transcript and I have to say deep gri it's just a pleasure to work with there's no rate limits it's super fast it's like it's like 10 seconds for one hour podcast it's it's unbelievable just amazing

I love hearing that and it's also awesome to have you as part of the startup community that we are rocking and I will say right now if anybody else out there listening would like to join the startup Community please reach out or just go to deep.com up and fill out the form we are open and we are accepting people still

dude Edward it's been a pleasure I really thank you for coming on here and doing this podcast with us I feel like it's a little bit meta we are having a podcast talking about podcast AI and later on I think we're going to be creating some podcasts using the very same AI technology that our guest is working with.

"WEBVTTKind: captionsLanguage: enall right welcome everybody we are here with Edward who is the creator and CEO of podcast AI a Toronto native went to the University of waterl but did not go for computer science went for biology because he already knew how to do whatever he needed to do with the computer Edward it's great to have you here and I'm looking forward to dive into what you're doing with podcast AI how you found building with AI in general to be but before we get into any of that I'd love to hear a little bit about your background and what you've been up to until now yeah uh pleasure to be on the show um so uh you know I you know born in Toronto uh from fairly early age like in the late uh '90s uh doing websites you know learning how to rate apps for for Mac uh for Windows you know it's it's kind of funny like in software these days everybody's talking about already front end back end you know I was doing stuff pre that being a thing um and you know it's been pretty exciting just seeing how software is kind of evolved uh did a lot of medical apps um so I was doing you know some Consulting doing uh medical apps and Canada um some n nationally used apps as well um and in 2019 me my best friend Johan duncomb we you know banded together uh he was my best friend since grade nine and uh you know we we did our first startup that was sort of in the Creator space um but then everything changed with you know we had heard of of uh you know G GPT and and logged into open AI for you know playing around with gpt2 you know it was kind of interesting you know it would you know you you'd give examples of YouTube titles and it would hallucinate more of them you know type thing it was like kind of useful but you could see how it could become useful but it's really when Chachi BD came out that change of the game and shaan and I you know we were like okay pencil's down stop everything we're doing we have to do this like this is this is the biggest thing we've seen in 25 years 30 years time to Pivot yeah so you were in the Creator space you're now still kind of in the Creator space with what you're doing with podcast AI can you break that down a bit yeah so before we're we you know our previous startup was uh doing something where it was trying to combine let's say a YouTube and and a and a patreon together and it's a very different business because it you know we learned so much we learned that you can't fight Network effects um you know that that's really one of the big lessons and and there's a lot of other lessons um that we learned um you know especially wanting to build a venture backed company and being in Toronto you know realizing that you know the value of ground game um you know hence we're we're we're working on moving down to California where where everything's happening um but yeah we we you know it's it's kind of amazing how fast we built up Podcast AI uh we only started in May and like we built this thing at break next speed it's you know the best piece of software I've written um and yeah AI is just such a big part of that in every way a culmination of all those past events now has led you to writing the best software of your life exactly all all roads are converging here yeah well break down some of the stats what have you got for us as far as users and what you're doing maybe GPT calls or transcribed hours all that fun stuff so we get an idea of what you're working with so uh we have let's say we have 25 customers right now paying customers uh we we we got our first customers in September um we took uh 15 no we took 19 sales calls in September October 15 of those we closed on the call um which was pretty bananas that's a good signal yeah and and um you know to to to sort of back up we we got into this by building in public so originally it was me playing around with uh the 11 Labs API and uh essentially creating a parody of the all-in podcast so I tweeted that and uh you know 10-minute parody and then and the sort of background idea was that the cast of Allin was reborn as AIS um they had ended an episode joking about like or started one joking about oh you're going to become an AI or something so I basically made that um it went viral it got retweeted by them uh I did six in total um but going through it people were like is this real like you know half the people thought it was real and half of them realized it was me scripting the show um and using their voices um but we realized like hey like let's build a company right now cuz what was happening with AI was just absolutely insane you had the momentum yeah we had the momentum like just the attention the second second time founder you realize distribution is everything you know you can have the best execution but if there's no distribution there's nothing um so we you know we realized uh we got an amazing name podcast AI uh I I think you can't really do better um for what we're doing now it's just you up one night like I wonder if it's available that that was Sean um we were we were looking for domains uh uh you know related to the other company and and um Sean had found the service that was you know G giving these domains for sale and we were just talking about this idea of doing an AI company uh around podcasting and you know I forget which one of us were like try podcast Ai and boom it was available for sale we bought it like on the spot hundreds of thousands of dollars later uh not even that much I think it was underpriced wow it was it was 10,000 no way I I think a steal I think a steal yeah yeah yeah yeah especially for doing what you are doing which is so clear for podcasting using AI I love that all right sorry I derailed the story but keep going well no no it's exactly that it's it's you know the the we decided okay we're going to build a company that is going to automate podcasting you know the the plan the mission is really automating everything a toz making it save human producers um hundreds and hundreds of man hours or person hours um and you know that that's that's the mission and even even creating like getting to the point of creating synthetic uh content we can generate ad read in the voice of the host ad libed about any you know sponsor and call to action and it's it's amazing um yeah there's nothing like showing that demo um but uh but so yeah like like it's it's a really ambitious um project um and uh we got into Jason calacanis is uh first his founder U program uh we got a small investment check through at the end of that 10% of the companies get that check and then we got invited into the accelerator program um early September and uh you know the rest is history so was that because of these viral tweets be honest and that you you buttered them up with well that's the start you know you have to get on people's radar that's the thing with like raising VC and that's that's the thing we didn't understand the first time in 2019 when we late 2019 when we started the previous startup it's that you have to network yourself in and you know you know VCS they just have so much incoming there's so much noise that you have to introduce some signal and get on the radar and you just have to impress them with something show them something amazing you know and that that's the secret yeah and especially when you're building with AI because there are so many companies out there right now doing something in this space that to differentiate yourself is a huge value when it comes to VC's looking at you and saying oh yeah I've heard of them because I've seen some viral tweets or I've heard of them because I've seen what they've been doing they've been building in public as you say and I really like this notion of building in public how have you been taking that like are you going as far as sharing all of your customers sharing your Revenue numbers all of that kind of Indie hacker building in public or is it more like hey check out this cool feature we just threw this out there does do people like it do they not screenshots of cool features um be you know part of it is is the product we're doing so for example uh we'll turn a podcast into a website um and the website will have you know statistics on the podcast all the links like you know so for example um this weekend startups that's like 1,800 plus episodes by Jason calanis uh we now power this weekend startup so that's just a beautiful demo of what we can do um all the transcripts are in there uh powered by you know deep G um there we uh you know table of contents that we generate all that great stuff all the distribution points all the advertisers we automatically detect the sponsors and you know rotate those through the transcript as like calls to action and finally an AI chat with AI versions of the host trained on all the past EP uh so you're chatting with the host yeah break down a little bit more about what podcast AI does so that we can understand all of the different features yeah so we eventually want to go a to zon podcasting I mean it's in the name podcast AI um but you know if you start there's sort of reproduction which is all the scheduling stuff so everybody showed us the spreadsheets that they use and we want to do a product for that um and then there's production which is the actual recording which is frankly a commodity you know there's Zoom there's there's tons of ways to record a podcast where we start right now is the postprocessing post- production meaning you've recorded the MP3 or or the video file um you upload it into podcast Ai and then we generate the description the title automatically put the episode number generate the chapters do the whole transcript speaker identified again thanks to deepgram we gener at the key points you know key takeaways for the podcast all those show notes type things um we you know we we generate right now viral moments from the episode top 10 viral moments um and we're actually doing now we just release this generating blog posts that go with that Viral moment um including LinkedIn post uh Twitter xread Instagram Tik Tok YouTube short um just generating everything and the the road map is actually to be able to post those directly onto the social accounts so that we really are automating the pipeline A to Z yeah saving so much time I do know that that is one thing that a podcaster spends a ton of time on is the just the data transfer and taking the video the edited video downloading it then uploading it to YouTube uploading it to Spotify or your podcast Distribution Center then adding the description and the description for Spotify is a bit different than YouTube because YouTube you can do different things with it adding all of the chapters adding all of the links and the different things that you talk about so it's show notes in that regard are you cuz I think this is a really hard problem that I would love to dig into are you automatically grabbing links so if we say hey you know what is really cool is the way that Tim Ferris does show notes and adds links and you can see that on tim.com podcast do you automatically pick that up grab it and then throw it into the show notes because that seems like it's a really hard problem to do and I would love it if somebody did that so that is literally a top feature request right now and we started work on it so uh you you will probably see it in the product than the month well realistically how can you make that happen because a lot of times like I just made it really clear for the transcript to understand what the link is but if I was saying something like oh you know who does podcast show notes really well is Tim Ferris you would need something to go out and find Tim Ferris podcast URL and then show different show notes from his podcast URL which seems like a very hard problem and you may need some kind of autonomous agent in there how are you looking at like attacking that type of problem so so it is hard and it isn't so basically with with AI the crazy thing is how easy it is to do something amazing in one call to the llm but to do anything beyond that what you really want to do is layer calls to an llm so you don't quite need an autonomous agent but what you need is sort of a procedure which is really doing what humans do right like we'll take one pass at something and then a second pass and a third and a fourth and a fifth and that's how we refine work products the you know the the app itself the back end really needs to be doing the same thing so what we would be doing is first pass detecting you know what is possibly a link for the show notes or intent for something to be a link in the show notes and then on a second pass actually attempt to retrieve that from the internet and determine what the link is so through a few passes you can actually do stuff like that and and you know it's just hard because you have to do it right it's not as easy as a onshot LM call but imminently doable yeah I imagine it also you have to think about costs and how much is this going to add to the end user experience this this this case I think it's very clear like this would be incredible it's obviously one of the top feature requests that you have but then on other features you have to weigh out those pros and cons like is this worth another llm call is it going to add that incremental value or is it going to be like hitting it out of the park for the end user experience and so it makes it worth it to add that 0.03 cents for the per token or whatever it may be whatever the cost is by the time this podcast comes out because it's constantly changing but how do you look at that like how do you think about those tradeoffs on the feature that you're introducing but on the back end the amount of cost that it's going to cost you and it's almost like your cost of goods or cost of services is going down that profit margin that you have is going down yeah so so the way we approached this when we started is you know in theory the best way to price your product is to determine the value and then you price according to the value um but to keep it simple when we started we just said okay we're going to figure out what our costs are we're giving on each subscription a number of credits and every action that costs us credits costs us you know money um we basically deduct from credits that you're you have to use right so we basically make it so the user is budgeting for themselves and deciding okay what's valuable and I think that those show notes will be valuable and also our plans are pretty generous on the credits so I doubt anybody would be bumping against the limit it's really just you know a guardrail so there's that and also like look we we're approaching this as we want to be the premium product in podcasting podcast AI um in the same way that open AI is the top you know series of of models that you can use use for llms so that's our intention and you know do doing that work and and here's something that's similar to do right um and similar paino in terms of intensity for users um guest research that is labor intensive we can automate that in the exact same way so that kind of becomes that you know pre you know pre-recording step that's you know right next to the scheduling that's also something we intend to do yeah the amount of steps that is involved in a podcast when you break it down it is not like any of them are very difficult to do but they are just tedious and so you saying you know what it's we don't need a genius to do this we could probably automate it with a bouquet of zapier and llm calls and maybe some chain together llm calls however you're doing that and some uh transcriptions right some calls to the different apis like deep gram and so it if you are able to do that and take you're giving me back time as a podcaster you're really opening up my ability to say okay now I don't have to worry about the schedule or the research I just got to show up and that's one of the things that I am a big proponent of is that with the podcast I want to figure out a way to make the process so simple that I just have to show up because that's what I enjoy doing right I enjoy talking to people and so if that is the end goal I'm all for that and I'm really excited for what you're building yeah 100% basically reclaim so uh again J Jason had tweeted something like I'm reclaiming my time I'm going from Seven podcasts a week down to four um and that that kind of when we were building the company that sort of became our slogan like let's let people reclaim their time and turn them into 10x producers so yeah you know that really is that really is the thing and and with podcasting it's like there are a lot of steps but the truth is podcasting can be as Bare Bones or as high production value as you want so for someone who really wants Bare Bones you know record zoom and flip it to YouTube that's it right you don't even have to describe anything you just put a title you're you're done but that's you know that's not going to do well what what you want is the production quality you want chapters you want summaries descriptions you want all of that good stuff you want a nice show website that's completely automatic you want to get the word out yeah exactly so you really have to you know up the game and and that's what we've kind of been seeing with podcasting right everybody's upping their game and the idea is for at least for the least work possible make it so someone can up their game with podcast AI you know up to that top level well man I appreciate this I want to leave with one final question because you are building in the space and AI is so new you're doing all different kinds of things with AI whether it is transcriptions it's or it's so you're kind of doing the whole pipeline I would say not just when you look at the podcast pipeline but when you look at there's audio coming in it's being transcribed you're taking insights from that transcription and then you also have the ability to create audio on the other side of it so you can go text to speech and so for me that that's basically the whole Gambit now what are some of the main challenges that you have faced while building with all of these different new technologies so the technology itself is unbelievably simple to use I'll just give you a few friction points that might have happened that did happen so one was you need a vector database and you know because you need a vector database um we had to switch from MySQL to postgressql um so that was a little database change that happened very early on because you get PG Vector which lets you um do those Vector searches so that that was one thing um the second uh big issue was with open AI getting access to rates like tokens per minute and stuff like that that was super important for us getting access to GPT 4 really early on was super important to to us and it's sort of hard to wiggle your yourself into into that thankfully we recently got a really big uh limit increase but yeah that was that was a big pain Point um but it's interesting how the actual working with the llms is just such a pleasure and and probably the literally the easiest one is deep gram because nice you know for podcast AI everything starts with the transcript right so you're uploading that episode and then you're hitting transcribe or it's automatically doing it for you and that's the basis on everything because every other llm is fed from that transcript so the fact that deep gram does the speaker identify identification that's like absolutely critical um and that lets us rapidly identify the speakers the hosts and the guests um and then everything else is from that the the chapters are done from the transcript the summary is done you know from from the chapters um so that's like a little a little cheat to to to go faster um and and and so so really the root of it is is the transcript and I have to say deep gri it's just a pleasure to work with there's no rate limits it's super fast it's like it's like 10 seconds for one hour podcast it's it's unbelievable just amazing I love hearing that and it's also awesome to have you as part of the startup community that we are rocking and I will say right now if anybody else out there listening would like to join the startup Community please reach out or just go to deep.com up and fill out the form we are open and we are accepting people still so dude Edward it's been a pleasure I really thank you for coming on here and doing this podcast with us I feel like it's a little bit meta we are having a podcast talking about podcast Ai and later on I think we're going to get to see you using podcast AI on this very podcast so it can't get more meta than that and we will end it there awesome it's been absolute pleasure to join you here and uh yeah I I love what you guys are doing and uh you're you guys are the backbone of of transcription I think\n"