Dexa goals to get extra out of podcasts with AI-powered search

In the event you hearken to plenty of podcasts, there’s a probability you would possibly bear in mind humorous tidbits and are questioning… “Wait, who talked about consuming fries with sriracha once more?” or extra critical questions. To search for the solutions, it’s a must to first discover the podcast after which search via their transcripts. Dexa is attempting to make podcast search simpler by leveraging AI.

The device helps you to ask questions on a single podcast, like Andrew Humberman’s Huberman Lab podcast within the screenshot beneath, or question all of the podcasts in Dexa’s database — there are at present greater than 120 with extra being added. The search outcomes will provide you with an AI-generated abstract of the reply together with tips that could podcasts the place the participant mentioned the subject.

As an example, you may ask questions like “What’s one of the best ways to get extra sleep?” and discover solutions to that from Dexa’s podcast library with timestamped hyperlinks to these conversations. You may as well @point out a selected podcast to slim down your search outcomes.

Picture Credit: Dexa

Once I requested Dexa “What does Mark Zuckerberg take into consideration Threads?” it returned the outcomes of his dialog on Lex Friedman’s present concerning the social community. But it surely additionally confirmed me outcomes the place podcast hosts had been speaking about Mark Zuckerberg’s remark about Threads on one other platform.

You may as well share these searches and outcomes with others via hyperlinks. Due to this, Dexa is ready to additionally index among the content material for engines like google like Google.

The corporate

Dexa was began by Riley Tomasek in 2023, who beforehand co-founded an app known as Flight for customers to seize notes, photos, and concepts on totally different boards and collaborate throughout them. The corporate was acquired by Figma in 2021.

Tomasek mentioned that he actually acquired into well being. As a part of his data constructing section, he listened to the Huberman Labs podcast. He needed to take a complement Huberman prompt however forgot the dosage. So he began looking out again for when the neuroscientist talked about it and the method took a very long time.

“After this expertise, I made a decision I’m going to construct one thing to make this simpler as a result of I don’t wish to undergo this course of once more. So I transcribed all of the episodes and used instruments like OpenAI’s embeddings and Whisper to make the search work and posted it on X,” Tomasek mentioned.

After Tomasek posted about his instruments Huberman retweeted it and that generated inbound curiosity from podcasters and listeners for the instruments’ growth. He mentioned that plenty of podcasters talked about individuals not having the ability to discover particular content material.

The startup makes use of totally different indexing methods combined with a data graph of individuals, episodes, and exhibits to energy its search. Tomasek mentioned the device launched practically 9 months in the past, and has answered over 1 million queries with 50,000 individuals visiting the location every month.

Dexa has raised $6 million in seed funding led by Summary Ventures and The Common Partnership with participation from Maple VC and Guillermo Rauch. The corporate at present has three individuals on the crew and it now plans to rent extra engineering workers.

The corporate is launching a redesigned website and asserting a partnership with Huberman Labs the place Dexa’s search might be embedded on the podcast’s website. For that, the startup has skilled a particular mannequin — consider it as a GPT in ChatGPT.

The brand new design lets customers ask follow-up questions and lets them watch the linked dialog with out having to depart the window.

Future roadmap

Dexa’s progress has principally been natural with podcasters referring individuals in direction of the device. He mentioned that this methodology nonetheless has momentum as the corporate works with well-liked podcasters. So it’s not seeking to put money into outreach for the time being.

“We briefly tried promoting on podcasts, and discovered that our makes an attempt led to a relationship the place we listed their content material, added it to Dexa, they usually began to organically discuss it,” he mentioned.

The corporate isn’t at present making any cash, but it surely has thought of each subscriptions and adverts as sources of income because it scales.

Dexa additionally plans to construct a cellular app, the place it could drive extra consumption to podcasts via quick clips. Moreover, the corporate is planning to introduce account creation so individuals can save their searches.

Alternatives and challenges

There are many instruments together with Riverside, Podcastle, and MusixMatch that assist podcasters with transcription. When it comes to search, the newly pivoted Very Disco is attempting to grow to be a kind of IMDB of podcasts with info like books, individuals, and TV exhibits talked about in podcasts.

MusixMatch’s chief product officer Marco Paglia sees the promise in what he calls the “assistant-driven UX” of Dexa. Plus, he feels that there’s potential in permitting individuals to retrieve a selected individual’s opinions via podcasts. However his purpose for not embedding such a search in MusixMatch is that he feels textual content isn’t one of the best ways to work together with LLMs.

“I really feel that typing on a chat isn’t the perfect UX for assistants. As an alternative, for this use case I strongly imagine that voice is the long run,” he mentioned.

Ben Cmejla, companion at The Common Partnership, believes that Dexa’s method of utilizing data from creators for search is exclusive.

“A platform’s success or failure will hinge on the belief it builds with customers and content material companions. Dexa’s method places belief first — consumer belief that solutions to their questions come from human consultants, and companion belief that their content material might be showcased, not simply used for coaching information or anonymized summaries,” he mentioned.

Leave a Comment