42 Labs - personalized dynamic podcasts

Technical Design
Select the Topic you want to learn about
Tell us whats your skill level
Select your preferred language for your podcasts for the topic
Select a preference of sources (otherwise we choose defaults)
We recommend smart sub topics that you can choose from to make sure the podcast reflects your learning goals
Dive Deeper to generate more similarly topical podcasts
Skip for topics you know already or aren't interested in learning about
Check out your personalized dynamically generated feed!

Inspiration

Learning on the go is hard. Especially when you are in the exploration phase of broad themes, and are bounded by the time you have to spend sifting through the internet. Imagine this, you are driving to work, stuck in traffic, and you see a bill board about "Vector Databases". Being in tech you are curious about what this means.

Let's take another step forward, and imagine kids in rural India who speak Maithli trying to learn from TED talks in English. Imagine a technology that synthesizes high quality, tailor made content from languages around the world, in their native language. Image the opportunities this access would create!

What it does

To solve these challenges we have created a personalize AI app that figures out your learning style, is able to intelligently query thousands of high quality videos, TED talks, articles and podcasts for around the world, to synthesize customized podcasts for you, the kids in rural India and anyone looking to learn something new.

How we built it

link We have built a native app for personalized, customizable, dynamic podcasts.

We take as input user parameters such as topic of interest, and comfort level with the topic i.e. beginner, intermediate, advanced.
Based on these, we search a few selected databases TED talks on YT (filtered by view count) and podcasts and index them using twelve labs generate index API.
Once we have the index, we then also prompt the user to interactively select chat gpt curated subtopics from within the selected topic.
This creates a customized feed for the user with podcasts across several dimensions of the topic selected. For example, if the user selected "Leadership", we will have in the feed podcasts about "leadership style", "team building" etc. This allows the user the interactivity to "skip" the podcasts (subtopics) they don't like and "dive" into ones they do

The eventual goal being to also take as input language and continue to evolve the personalized user learning profile with the feedback we get from topics requested as well as the "skips" and "dives"

Challenges we ran into

Stitching together several systems across chatgpt, eleven labs and twelve labs wasn't out of the box.
Connected synthesized eleven labs outputs wasn't straightforward
We spent a lot of time to curate the right types of prompts to generate effective video summaries from the twelve model
The backends across several models were slower than we anticipated. There is potential to quantize these further and make them faster.

Accomplishments that we're proud of

The final product is something each one of us sees ourselves using! We were able to come together and light up an end to end experience in a few hours with easy to use APIs and powerful foundation models from eleven and twelve labs alongside gpt 4. The potential of making knowledge accessible to those with language and learning barriers with a product like this is a leap that is fascinating!

What we learned

We leant a lot on the way!

How to figure out innovative ways to make AI systems talk to other AI systems
The potential future in intersectionality of varied content
The exponential space that adding language opens up in dynamic podcasts

What's next for 42 Labs - Personalized Podcast Builder

In theory this idea could be extended to creating not just podcasts but also learning videos - long form and short form. We would love to make AI creators personalized for users.

Checkout demos here! link