Casey Traina, CO-Founder & CTO at Overlap AI. Overlap AI is building AI agents that can search, edit, and understand video content for media companies, podcasters , and creators. Their first agent transforms video marketing and can transform long videos into short form content, ready for distribution, and in your specific style. They were part of Y Combinator in Summer 24 and closed their seed round in November of 2024.

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

In this episode of the AI Minds Podcast, Casey Traina, Co-Founder & CTO of Overlap AI, shares his journey from a young programmer in Philly to building AI agents that transform video content creation.

Casey discusses his decision to leave Stanford, his early entrepreneurial drive, and how Overlap AI evolved from a podcast discovery tool into an AI-powered platform for automating short-form content.

He explains how Overlap AI enables creators to seamlessly clip, edit, and distribute videos in their unique style, leveraging AI to learn from social media performance and optimize engagement.

The conversation explores the challenges of AI understanding video, the complexities of automating high-quality content, and Overlap AI’s vision to elevate production standards and expand its capabilities.

Casey also shares insights on the future of AI-driven content creation and invites listeners to test Overlap AI by joining the waitlist or booking a demo.

Show Notes:

00:00 Innovative Leak Detection Business

05:11 Founding Trellis: Streamlining Data Processing

06:56 Automate PDF Workflows Pain Point

09:46 Trellis: Traceable PDF Data Processing

15:43 "Persisting Challenges in PDF Automation"

16:45 "Automated PDF Processing with AI"

More Quotes from Casey:

Demetrios:

Welcome back to the AI Minds Podcast. This is a podcast where we explore the companies of tomorrow being built AI First. I am your host, Demetrios, and this episode, like every other episode, is brought to you by Deepgram. The number one speech to text and text to speech API on the Internet today. Trusted by the world's top enterprises, conversational AI leaders and startups, some of which you may have heard of like Spotify, Twilio, NASA and Citibank. I have the pleasure of being joined by the CTO and co founder of Overlap AI Casey. How you doing today man?

Casey Traina:

Doing well, thank you for having me.

Demetrios:

Sweet man. Well, it's great to have you here. I want to know a little bit more about your life and what you've been getting up to. Before you started Overlap, I know you grew up in Philly, so go birds. And then you decided to go to Stanford and were there for a little bit, but for some reason it didn't stick. What happened? And walk me through the transition and what you did to eventually drop out.

Casey Traina:

So I started programming when I was pretty young. I started when I was 10 doing Swift, which is like the iOS app development language. I was building a bunch of apps growing up, doing some research, kind of any ad hoc work that I could find. And it was there that I kind of fell in love with the idea of building and engineering and innovation. And that one ultimately led me to choose Stanford and pursue CS there. And then while I was there, it was amazing. I love Stanford. I would give my life for the school.

Casey Traina:

It's amazing. But there's definitely Or at least I personally had a bit of a misconception between the relationship of computer science and building and software engineering. And I found myself doing more math and more theory than I was doing application. And so I still had this, this itch that wasn't being met and it was to build new things. And I met my co founder at the end of my freshman year. And initially we had this value prop that we loved podcasts and there aren't enough or there are too many out there and there are too few being served to us. So we started that we really wanted to make a short form podcast app.

Casey Traina:

And then as we went down that rabbit hole and started building that out and working with more podcasters, we had this need that was pulled out of us that was for the technology that would power a short form podcast app and that was creating the short form content. And that's kind of where Overlap came. So now we help build these video AI agents that can search, edit, and understand video content for creators who are looking to create social media ready clips in their style and without them doing anything.

Demetrios:

So you got to Stanford and you were like, wait up. Am I in the wrong building? I'm doing math all day. I wanted to be creating things. I may have chosen the wrong major. And there's a big jump, though, between not enjoying your major and then saying, I'm going to go all in or something that I think is worth materializing in this world. Can you walk me a bit through that transition?

Casey Traina:

I mean, first I would say I loved my major and I loved all my classes and everything about Stanford, but I saw the world of AI and innovation kind of speeding by and I was worried that I would miss it. And I felt that I could. Could contribute to it immediately without waiting. Waiting the four years. And then I went to a talk with Sam Altman at one of our. It's called Entrepreneurial Thought Leaders. It's like one of our speaker series classes.

Casey Traina:

And he said his advice for anyone would be to drop out and to start a company. And I was kind of right in the crisis of deciding if I was going to leave or stay. So I kind of pushed me over the edge.

Demetrios:

It was like he was speaking right to you.

Casey Traina:

Exactly. We locked eyes for a second, I swear.

Demetrios:

I bet he doesn't get invited back after saying that kind of stuff. Stanford's like, dude, you just lost a ton of revenue. That's hilarious. Well, how did you go about finding the idea? It was just that you recognized you really enjoyed podcasts. You did not have enough podcasts on rotation that you felt like were serving your needs. You knew there was more than just Joe Rogan out there. And so then what happened? Like, walk me through the product creation.

Casey Traina:

So we started out with this idea that there were millions of podcasts in this world, but there were such a very. Such a few subset of was being served to the broader public. This is like Joe Rogan. It's like Call Her Daddy is like household names of podcasts that are rampant, but the other millions of podcasts which hold valuable information aren't being served. So we had this initial idea that we wanted to democratize the podcast industry and the way to do that would be to do short form so that people are no longer looking at the creator's name, but they're looking at the content and the virality that it's created.

Demetrios:

Like a TikTok for podcasts.

Casey Traina:

Exactly, yes. And while doing that, we were Approaching podcasters, asking them to work with us to get permission for their content, stuff like that. And the more we worked with them, the same question kept popping up of like, how can I get these clips that you guys are posting? Our app was full of like one to four minute clips. And how can we get these? These are better than our editor is making. These are phenomenal. We need them. And so at some point we were just emailing them to people and we were like, this is probably not the best way to do this. So we built a product of U1 and it kind of just took off.

Casey Traina:

It was immediately sticky with podcasters. People loved it. And there are a ton of clipping tools out there, but for some reason, people just were excited by the output of ours. They said it was better, it found the moments, it didn't interrupt speakers, stuff like that. It just felt more coherent. And so we went all in on building that product out to build a fully autonomous agent with social posting and learning and iterative feedback kind of all built into it.

Demetrios:

What does learning need?

Casey Traina:

So learning is when you create a social post, it will track its virality, and over time, it will learn to pay more attention to posts that have done well or have traveled far across socials, and then it will inform itself when going to find new clips.

Demetrios:

How cool is that? And so then you can ideally have things that go viral more often that.

Casey Traina:

And social media is a quantity game. So if you are posting more because it's automated and you're posting better because it's learning, then then you're going to win or you're at least going to 100% do better than you would have on your own.

Demetrios:

No, the fact that you were emailing these to people so that they could then go and post them on TikTok is an interesting sign. You're maybe our idea of a podcast, TikTok, isn't the right thing. Let's pivot. And you recognize that real early. So you weren't attached to any one specific idea. You got traction with podcasters on this. And like you say, there's a ton of different clipping tools that are out there. I know that just in the editing tools themselves, you have clipping tools.

Demetrios:

And so how do you look at it now as you're thinking, when this product, or as we build more for this product and mature the product, what are you doing to keep that your users happy?

Casey Traina:

I think the approach that we're taking is to build an end to end agent. So it's more autonomy in this. A podcast producer or a podcast host who is worried about who they're going to bring on or how they're going to prepare for their next episode. Stuff like that doesn't want to have to log on to Premiere Pro or capcut or something to edit their own clips. Nor do they want to hire someone who's going to cost tens of thousands of dollars per month to make these clips. So what we do is we have a completely verticalized end agent that will link up to your YouTube channel or your Dropbox or your Google Drive. It will listen for a new episode and the second that episode is launched or is released, we will take it, make clips from it, add your personal subtitles, watermark, music, whatever your personal branding and styling is, and then we will email those directly to you.

Casey Traina:

And you just have to hit one button that is post and we will learn from your language that you've posted previously and prepare that the copy for it immediately to launch on LinkedIn, Twitter, TikTok, et cetera. Instantly amazing.

Demetrios:

Just so more than, hey, we'll give you this clip.

Casey Traina:

It is.

Demetrios:

We'll take all of these steps and do them for you. And I know as a podcaster, it's not necessarily the identifying the clip that is the cumbersome part. That's the fun part, to be honest, because you're finding those moments where, this really resonated with me. I loved when the guest said this or that. And then everything that you need to do once you clip that, from exporting it on whatever it is, Premiere Pro or Cap Cut or davinci, to uploading it and then putting the tags and then trying to craft the creative around it. It's so cumbersome and it takes a ton of time. And then I'll go into tools like if I'm uploading it to YouTube, I'll go into tools like Vidiq and try and optimize it through there even more. And so it's multi steps and if you can, if your value prop is then like, we're going to eliminate all of these steps and we're going to do everything that you normally do, just better and more.

Demetrios:

I'm on board.

Casey Traina:

That is our goal. And you could still interact with it like you would a normal video editing agent or video editor. So you can still go in and use the chat. If you remember, there's that one really funny moment where Demetrios says X to Casey. you can go in and you can ask it to clip that moment and it will clip the 30 to 2 minute segment of it and it'll verticalize it and then also make a horizontal version, launch that to different socials and it really just supercharges your accounts because it's a quantity game and no one fully understands it. Even the companies themselves don't fully understand it. And so the more you can get out there and learn from what is doing well, the better you'll do over time.

Demetrios:

So I also write blog posts with the videos and the podcasts. Can you do that for me too?

Casey Traina:

So when you sign up for an account, with your permission, once you log on, we'll look at all of your previous posts and then for each platform you have a separate social media agent which is the one that creates all of those posts for you. So if LinkedIn, if you're doing blog posts, you'll get a blog post and if TikTok, you're doing just hashtags, you'll get just that.

Demetrios:

What are some challenges that you've had while building with AI?

Casey Traina:

The hardest problem is that video is really clunky and it's really hard to conceptualize how humans perceive video. So I always say that video is the unsolved medium of AI. That's why you can't drop in a video to ChatGPT and spit out whatever you want. And that's why generative video still is kind of in beta phase and like 30% done at best. Because video is really hard to understand because of that. The clipping algorithm, learning from what's doing well in videos is really difficult. Even something just as simple as question and answer, like what's happening in this video? GPT4O, I think only has 39% success rate on that.

Casey Traina:

I think it's the benchmark it hit and that's the state of the art. So absolutely. It has been adapting videos in a form that LLMs can understand and various different legacy models. And I think that was something we were wrestling with for a few months and it took us a while to crack and I think we now have it. So I'm excited to see where the product goes over the next few months.

Demetrios:

What else do you envision being able to do because you have cracked the code there?

Casey Traina:

For us, our two to three month vision is to up the production quality of our output because right now there are junior editors in the world that are just like adding subtitles to a video and then calling it done. And then there are senior editors where you'll see really highly produced videos where subtitles come from behind people's heads and there's like adaptive color grading and stuff like that. And that's where we want to get. because now that we can analyze frames and we can actually understand the video itself, then we have the ability to script that technology into it.

Demetrios:

Well, I'm pretty sold on it. I want to at least give it a test. What is the best way for me to do that? Just to go to the website, drop in a video.

Casey Traina:

So anyone who wants to try out the product, you can go to Overlap AI. You can sign up there for the waitlist, book a demo, or you can email me directly CaseyVerlap AI and we can book a time to get you on the product.

Demetrios:

Props to you for grabbing that AI domain. I didn't know they still made those.

Casey Traina:

Got it done.

Demetrios:

They were all taken by.

Casey Traina:

Any single word AI. It's hard to get.

Demetrios:

Very hard. So congratulations.