Sam Sethi: 0:14

Hello and welcome back to Podnews Weekly. I'm joined today by Joe Tannorella. Joe is the CEO and co-founder of a company called Pod Engine Tell. Joe, hello. How are you?

Joe Tannorella: 0:27

Hey. So I'm doing great. Thanks for having me. How you doing?

Sam Sethi: 0:29

Very well indeed. Now, Joe, what is pod engine.

Joe Tannorella: 0:34

To pod engine is an audio intelligence platform. So what we're doing is audio mining podcasts at scale, and that allows us to extract deep analysis and insights across the industry as a whole, all the way down to individual episode level.

Sam Sethi: 0:49

That sounds really exciting. How are you doing that?

Joe Tannorella: 0:53

So using our experience in computer science and a basement full of servers as well as set the art models as well as open source models, we are extracting, first of all, we're transcribing episodes at scale. So many thousands of episodes per day. And through that, we're running our own pipeline of analysis. The we extract things like who's advertising, who's guesting, what's their bio, what might the audience who listens This be interested in topics and themes discussed sometimes a sentiment, a bunch of stuff, and that enables us to do some pretty cool things.

Sam Sethi: 1:25

So let's take Port News Weekly, which has been going for about three years. I can't believe it's that long already. What could you do with something like Poke News Weekly?

Joe Tannorella: 1:35

So right now, actually, Pod News weekly is one of the podcasts that we do the deep analysis on. So if you go to a pod engine, you'll see that and search for Pod News Weekly. When you look at an episode page, you'll notice that we write a full article about the audio, which is great for accessibility reasons as well as we highlight the key topics discussed. We're very specific with those. We elaborate on them, the key takeaways. We highlight who's appetizing if there is an appetizer, the topics discussed in a weighted fashion as well, who the guests are and the bios, a variety of information. And very soon that will all be searchable not only in the potential interface but also through our API.

Sam Sethi: 2:12

So let's break down a couple of those things. So first of all, you talked about right, writing an article, I assume that's the air you know yourself and your co-founder writing an article every week for us.

Joe Tannorella: 2:23

Correct. I wish I could write that quickly in high quality.

Sam Sethi: 2:25

Now what I mean, again, just because people will want to know, is it an Al-Alam? That's open source is an open API? Is the chatter obtained or is another one that we haven't heard of? What are you using?

Joe Tannorella: 2:41

Great question. And a little bit of everything is what I would say. We lean very heavily into open source. So I think I mentioned our placement of servers earlier. We actually have a basement of servers and graphics processing and everything else going on. So we do lean into open source a lot for most of our processing that we're required. We also lean into state of the art models and we use pretty much all of them for different reasons as well. So a good combination of them. And of course we run our own transcription as well.

Sam Sethi: 3:06

Now you talked about key topic, are they extract Ted by looking at the transcript and then you talked about weighting them as well. Tell me more about that.

Joe Tannorella: 3:16

Yeah, correct. Yeah. So we do our best to extract the topics and provide estimated weighting. And of course this data's is improving week on week at this point and very soon we'll be tying that to wiki data definitions as well. So to provide even more enriched data and insight.

Sam Sethi: 3:31

Okay. And then lastly, you talked about the fact that you're going to be able to have a holistic view across the industry. How are you going to achieve that?

Joe Tannorella: 3:40

Yes, I think it's really interesting. So today, down to the episode level, as we've talked about, we can look at topics, discussed, who's advertising, who's guessing and so on. If you go up a level, then you can look at the podcast, watch this podcast, talk about generally, and if you could extract trends there and use advertising generally and so on and so forth. Then there's a few levels up as well. So you could look at maybe the general level and maybe it could be the location of the podcast is, which is some data we've started to extract. And really at any level it could be genre, it could be and I think you can search and filter by we can provide a 360 view of that using the data that we're extracting from really the lowest level is the transcripts. But we bubbled up back up to the top level then.

Sam Sethi: 4:18

So today, from what I understand, to use iTunes similarities to abstract, similar podcasts to each other. Now we talked offline about potentially using a topic that James and I talked about something called RAG, which stands for Retrieval Augmented Generation, which is an AI process for linking similarities and transcripts in podcasts, is a very good use of that. What are you doing in that space?

Joe Tannorella: 4:47

Yes, correct. So at the moment we do use RAG in a few places throughout our estate. So just as an example of that, we were sending a free kind of curation and discovery email to our users. And the way that works was they could tell us what they cared about in terms of kind of topics. And then we would then each week look at all the transcriptions that we've processed that we can send to the best episodes in our opinion based on this topics. That's quite a basic view. So Rack is what most people are using when they're doing things like chat bots, for example. But the graph rack really takes it a step further because it's a way to connect similarity and identify similarity across, in our case, episodes and podcasts. If guest A is being interviewed on a podcast and particular brands being mentioned in that same context and it makes it easier to spot the connection of those two things on different podcasts. If episodes and so on and so forth. So yes, we're really excited to be looking at that because Microsoft, as I'm sure most your listeners know, actually did experiment and train on podcast data when they released it.

Sam Sethi: 5:45

Yeah. How does that get exposed within the application and is it a visual tool or is it a written tool?

Joe Tannorella: 5:54

So if you go to the website today, depending on when this is aired, you won't see it. But in the next couple of weeks, one of the first products with built based on our API, which is using all this enriched data, is a guest booking tool. So if you'd like to place yourself or a client as a guest on the podcast, our ability to surface insights and transcripts, everything else makes it much easier to hone in on relevant podcast more quickly. So today that works. You know, it's great. We've had really good feedback on it and some people have migrated for other talking to us, which is fantastic. But to take that step further, something that people are really interested in is they spend a lot of time writing outreach, emails. The podcast are not only more often than not, the emails bounce because no emails they are correct, but also is very hard and time consuming to write that email. So one thing that they're really interested in is, you know, we know the topics that they're writing about in terms of the outreach. We can say we can look through the last ten transcripts from podcasts and how you spoke about climate change two episodes ago. But actually we can find the white space as well as you spoke about climate change in the context of farming, but you didn't mention this specific point. By the way, my client is an expert in that specific point. How would you like it if they came on to the podcast kind of thing? So that's an area finding the white space.

Sam Sethi: 7:02

Good. Now, one of the other example probably projects or platform services that you've built is keyword search, which is the ability for maybe James or I would want to put a tracker on our names or on our podcast. I've already built that, if I'm right.

Joe Tannorella: 7:23

Yes, that is live today. And we have some very happy pain users on that product. It was actually the first product we built. So initially when we built Pod Engine, we had this shiny technology. We didn't know how to surface it to paying customers or what problem solved exactly. So the first proposition was, like I mentioned, tell us what you care about. We'll give you good episodes. And we realized that whether through ego or something else, people were putting in their own names that are brand names that can better teams into that service. And so we realized actually there's a position here for media monitoring across the podcast world. And so that's where we built the product initially. And so yes, you can go to website today, everybody gets one free key red alert, and that's across for transcripts. And so that's a product that is near and dear and close to our hearts is the very first one.

Sam Sethi: 8:05

Now, the basis of this is transcripts, and that's what you're using is the ability to audio mine against. Now, if a podcast doesn't have a transcript today, are you you're transcribing that automatically yourselves, correct?

Joe Tannorella: 8:20

That is correct, yes.

Sam Sethi: 8:22

Would you make those transcripts available to purchase for the user? So let's say I'm a podcaster. My host doesn't support the ability to create a transcript. I'd really love one. Is there a way that I could come to pod engines? Hey, I know you've got my transcripts for these episodes. Can I just have that as an exported file that I can use within my own RSS? Is that possible?

Joe Tannorella: 8:48

Yes, absolutely is possible. And that's available today. So if you go to the website today, search your podcast, and if it's being transcribed, which is quite likely that you'll see the transcript there. And if you log in, you can just download it. If it's not there, then shoot me an email and I'll be very happy to start processing it for you. We've actually had a lot of interest from podcasts because a lot of the problems that I think we're able to solve weren't really feasible to solve pre the sort of alarms and tackle scale. And so we've been asked to provide a number of other areas of functionality for podcasts. So for example, if you go to an episode page, you can chat with the episode in a kind of chat like manner. So we're going to provide that to podcasts to embed into their own websites as well. And along with that, they'll get extra analysis that you don't get today. So what questions are people asking of the podcast? And even on pod engine, when someone searches for technology, for example, like is your podcast, which is about technology, is that shown up, if so, alongside which other podcasts. So yes, we're building a proposition for podcasters today. Very soon we're going to formalize that and bundle it up. And at the moment we're taking feedback from podcasters. So what would you love to see? So knowing that we have this technology and the sky's the limit, what would you build with our tech? Yeah, that's our approach today.

Sam Sethi: 9:55

Okay. Now one of the things you mentioned was an API, which is the way that my developer would be able to use to access some of this information. When do you expect to have the API available for developers to get stuck in?

Joe Tannorella: 10:11

So we're actually onboarding customers at the moment and some of the use cases coming over to us are really interesting. And like anything, when you're speaking to customers, you learn and they move forward with things. Just as an example of that, while we're on the topic, we're working with a a kind of counselor who a US listed consulting company, and they're looking to extract insights about the sentiment around the social housing market pre and post UK changes to Labour government, which is obviously extremely niche, having started to look into this area for them and with them there's actually a significant amount of discussion happening on that very topic. Another example is a pilot that we're about to run with somebody who's looking to identify climate change misinformation across the podcast world. So who's misinforming people and what can they do about that, Really? Yes, the API is being used by us in production and from September onwards we're going to start to onboard more customers into that excellent show.

Sam Sethi: 11:04

This sounds super exciting. If I want to go and find out more about it, where do I go?

Joe Tannorella: 11:10

Head on over to Pod Engine four slash pod news and you'll get 50% off your first month tastic.

Sam Sethi: 11:16

Thanks, Joe.

Joe Tannorella: 11:18

Thanks a.