I Tried to Build a Podcast Index. Here's What Happened.
For the last six months, I’ve been quietly working on a side project I called AiPodFM. The idea was ambitious yet simple: to create a fully searchable knowledge base out of the top podcasts in the world. What if you could query the entire podcast space of spoken audio just like you search the web?
This project took me on a bit of a journey, and while I've decided to pause it for now, the lessons I learned about AI, product development, and user habits are worth sharing.
The "Why": Solving My Own Podcast Pain Points
As someone who listens to at least five podcasts a week, I constantly ran into frustrations that felt solvable with today's technology. My motivation boiled down to three key problems:
- The "Where Did They Say That?" Moment: You're trying to recall a specific detail from an episode you heard weeks ago—a book recommendation, a technical concept, a name. You know they said it, but finding that exact timestamp is impossible. I wanted a way to search the full transcript and jump directly to the right moment.
- Smarter Discovery and Summaries: With millions of episodes out there, how do you decide what’s worth your time? I envisioned an intelligent layer that could provide a "sneak peek"—a concise, AI-generated summary with key topics and chapters, so you could decide in seconds if an episode was for you.
- A Knowledge Base for Audio: This was the ultimate goal. Podcasts are a new medium for information, full of expert interviews, niche discussions, and breaking trends that don't always make it into text. Imagine trying to learn about a rare beer brewing technique discussed only in a small podcast—Google can't help you there. What I envisioned was, essentially, a "Perplexity for podcasts": transcribe, index, and make the top podcasts searchable and explorable like the web, not just for curious listeners, but for AI agents as well.
The Breakthrough: The Tech is Finally Here
When I first thought about this, it felt like something only big corporations could solve. But I quickly realized that with the rise of multimodal Large Language Models (LLMs), the technology is already here. These new AI models don't just understand text; they can process images, video, and, crucially, audio, and the costs are dropping significantly.
I ran experiments with models from Google and OpenAI, and the results were quite promising. The AI could process audio, extract topics, and identify key moments with impressive accuracy. The dream was indeed technically feasible. The biggest challenge wasn't the transcription, summarization, entity extraction, or indexing, but building a clean, fast, and intuitive user interface—a "wrapper" to hide the complexity and deliver the magic, together with some compute to run the models.
The Twist: I Found a Service That Kind of Does It All
About three months ago, I made a discovery: a service already existed that did most of what I had set out to do. The service is called Snipd, and I would recommend everyone who listens to podcasts regularly to try it out.
My initial reaction wasn't disappointment. It was joy. The tool I had always wanted to exist was real! I immediately signed up and started using it. I never really thought about turning AiPodFM into a business; the idea was more to have a project I could work on, which would solve personal problems, and to see that a company already did that was more like a validation of the problem.
Important Lessons After I Stopped Building
This is where the story gets interesting. Even though my ideal product now exists, my experience as a user taught me something in addition to what I learned from building.
- The Chasm Between "Cool" and "Willing to Pay": The service has a free tier that covers most of my needs. Despite being the ideal user who desperately wanted this solution, I found myself hesitating to upgrade to a paid plan simply because I don't need more than 5 AI summaries a week. It was a humbling lesson in product-market fit: creating something people want is one thing, but creating something they're willing to pay for is a whole different mountain to climb.
- The Immense Power of Habit: Even with this powerful new tool at my fingertips, I still sometimes find myself defaulting to Spotify—especially when friends send me recommendations. The muscle memory and network effects are sometimes just too strong. I have my playlists, my followed shows, and my routine. Breaking a user's deeply ingrained habits is a challenge, even if your product is objectively better for a specific task.
While AiPodFM is on pause, the journey was a success in its own right. It proved that the future of searchable audio is here, and it taught me valuable lessons about the gap between a great idea and a viable business. And now, I get to be a happy user of the tool I once set out to create.
Try It Yourself
The core idea lives on as AiPod, my podcast answer engine: ask a question in plain English and get a cited answer with timestamps pointing to the exact moments it was said.