Cover Image for Introducing our Intelligent Podcast Engine
Launch Announcement

Introducing our Intelligent Podcast Engine

By Pierson Marks

It’s been two and a half months since we released Jellypod on the iOS App Store, and our users have already created over 45,000 minutes of audio with Jellypod. Many listeners open the app every single day as part of their daily routine, listening to a summary of their personal news in a convenient audio format.

A Better Podcast Experience

Over the past week, we started to roll out our most exciting update yet - one that significantly improves the quality and consistency of your daily pods. By listening to our most active users, we discovered that almost everyone wanted a “smarter” podcast, where Jellypod recognizes similar topics from any of your newsletters and combines them into a single podcast section.

This would enable all perspectives on a subject (from many different sources) to be in one section, just like how a human news anchor would deliver the news.

We also wanted these sections to be arranged by importance but ordered in a way where related sections would be close to each other. Your pod should be a cohesive listening experience that doesn't jump around too much. For example, if one section talks about the Apple Vision Pro, and another on Apple’s financial earnings, those two sections should probably be next to each other.

Building a Novel Algorithm

To intelligently perform de-duplication and semantic ordering of topics, we had to create a new architecture that could efficiently process an unbounded number of input tokens - potentially even much larger than the current context window of GPT-4.

Even with advanced prompt engineering and chunking with GPT-4 (or other models), it was impossible to accurately and consistently detect and organize newsletter subtopics into a high-quality podcast with a deterministic length.

We needed to guarantee that your podcast always includes the most essential topics from all your news without repetition. However, consistently producing a deterministic and controllable output is difficult with artificial intelligence. And even in a future with GPT-5, AI just doesn’t have the context to know what actually matters to you as a listener, and there’s still a chance that model “laziness” leads to content that doesn’t meet our quality bar every time.

We believe it’s unacceptable to deliver a product where the output meets our expectations only 90% or even 99% of the time. Software reliability is usually measured in the order of nines, so why are AI applications any different? No one in their right mind would ever use a software product with only one or two nines of availability.

This same standard should apply to consumer AI applications.

So going back to the drawing board, we created a novel algorithm to power our new Intelligent Podcast Engine. We believe it is the first of its kind to combine multiple methods of semantic chunking, organization, and statistical analysis with an agent-based architecture to consistently output a properly ordered and deduplicated summary (independent of the base model’s context window).

The changes in your podcasts may be subtle at first - but with time, we guarantee you’ll notice a profound improvement in quality consistency, topic cohesion, and ad removal. We’re extremely excited for everyone to check out these changes and to hear your feedback as we continue to refine and enhance your listening experience.

What’s Next

I couldn’t be more excited about the future at Jellypod. I wake up every morning with a vision of a world where the content you consume perfectly reflects your interests without the sensationalism and fluff of traditional media outlets.

It shouldn’t be a chore to stay up to date with the news you actually care about. And in a world with more independent thinkers, writers, and publishers than ever, we must make these voices more accessible to everyone.

Jellypod will continue to push forward new ways to ensure you’re always in the loop no matter how busy life gets. In the coming months, we plan to roll out even more features, including direct integration into your favorite podcast player, in-app transcriptions, and much more.

Thank you for your continued support and for being a pivotal part of our journey.

If you are interested in joining the team to help support this mission, please reach out to