I Built My Own Podcast Intelligence System
A VC fund built a podcast intelligence system to discover companies before everyone else. So I built my own version—automated transcription, speaker ID, daily summaries with cartoon headers—in a couple days. This is what agentic data pipelines look like, and why they're the next wave of personal software.
Howdy y’all 🤠, and welcome to another blog post where I probably spent way too much time solving a problem!
I learned Tomasz Tunguz built his own system that automatically transcribes and summarizes podcasts to discover companies and talent before everyone else, so I built my own version in a couple days.
Turns out building information advantage is easier than you think.
A few months back, I wrote about building custom tools in 30 minutes with Cursor. Calendar analytics, running art and plan generators, etc. This podcast system is same idea but a bit more advanced: a production system I now consume daily. Credits no longer owed to Cursor, but Claude Code (literrally.. ;)).
Meet TEAHOSE.com
My fully functional podcast intelligence system (yes, I may have named it while slightly under caffeinated). Every day, this thing:
Presents new episodes from ~40 podcasts
Downloads, transcribes, and diarizes those I select
Uses Claude for speaker ID and summaries
Publishes to a newspaper comic-style website (Ty Opus 4.5 for design)
Sends daily email digests with fun AI-generated cartoon headers (Ty Nana Banana Pro, rip comic artists..)
The daily emails look like this:
The website has simmilar fun comic-style vibe:
Keeping Up With Joneses
I obviously can’t listen to ever episodes of 40+ podcasts. But I can scan summaries over coffee every morning, deep-dive what grabs me, jump to exact timestamps. The summary shares key themes, contrarian takes, companies and people mentioned, overlooked insights.
I get the All In tea. Learn why General Atlantic on why to invest in Latam contrarian takes, and Airtable CEO’s latest business ideas. Just some examples.
Oh, and there’s a chat feature for asking questions about episodes:
The Stack
MLX-Whisper: Local transcription on Apple Silicon
Claude (Haiku + Sonnet 4.5): Speaker identification and summarization
Nana Banana Pro: Cartoons
Vercel: Website hosting
SendGrid: Daily email digests
PostHog: Analytics
The admin workflow is just a Python script with an interactive menu. Run it once daily, select episodes, it processes for 5-10 minutes. Might have it run on the cloud and then text me the list daily for me to select. Could also be fun to fine tune a model to replicate my curation:
Build a PostHog dashboard to show me what’s resonating:
Future Musings
Building this got me thinking about agentic data pipelines, systems that autonomously collect, filter, and surface valuable information.
Personalized News Curation
I’m a big fan of News Minimalist, which does something clever: ChatGPT reads tons of articles and ranks them by significance. You only see things above a certain threshold. It’s really good at surfacing scientific discoveries you’d otherwise miss.
I can imagine a world where everyone has their own version of this, an agent scanning everything and curating based on your specific preferences in terms of content and writing style. That’s essentially what I built here, but for podcasts.
Alternative Data for Hedge Funds
Tomasz’s use case is really about building a data pipeline that generates alpha. Which got me thinking: what other types of information pipelines could be valuable?
A friend of mine mentioned a company that puts solar-powered devices near key energy transmission arteries. They cut deals with local landowners, measure energy flux at generation facilities, and sell that data to energy traders. Physical sensors and signal transformation creating proprietary datasets.
This podcast system is really the same concept. An agentic data pipeline that captures and transforms signal to create alpha. Quite a neat business vector.
Get the Tea
Feel free to subscribe at teahose.com if you’d like to get the tea too. Fair warning: not sure how long I’ll keep it running but for now, it’s live and the cartoons should make you smile! Also, if you do subscribe and read it, would love your feedback! All feedback is appreciated: good, bad, and anything in between.
As always, if you enjoyed the read, please feel free to share with friends.
And let me know your thoughts/reactions/questions on any of this. Conversations that come out of these posts are the reason I do them. You can reach me at altmbr@gmail.com or on twitter @altmbr. And for anyone new that wants to get future posts in their inbox 👇