How to digest 36 podcasts in one daily pass

TL;DR

The system downloads daily podcast files, transcribes them with a local pipeline, and produces structured summaries. It extracts topics, quotes, and potential investment theses, then drafts blog posts in the creator’s style, all via a terminal driven workflow designed for speed and personalization.

Transcript

I have a list of 36 podcasts, but I don't have 36 hours every week to listen to 36 podcasts. So, what I did is I created a system that goes through each of those podcasts every day and downloads the podcast files and then transcribes them. Can you show us how it's actually built? Like where do you get this feed? It sounds like you run it locally. H... Read More

Key Insights

• The Parakeet Podcast Processor downloads and processes podcasts daily to convert audio into usable text and summaries.
• The pipeline uses ffmpeg to convert audio to text and Gemma 3 to clean transcripts for better downstream processing.
• A local duct DB tracks which podcasts are processed on which day, enabling reproducible daily outputs.
• Daily summaries include host, guest, comprehensive summaries, key topics, and key themes for quick comprehension.
• The system identifies noteworthy quotes and translates them into actionable investment theses for venture capital contexts.
• Prompts are generated to draft blog posts in the creator’s own style, ready for further refinement.
• Despite a strong focus on local processing, the workflow prefers terminal based interaction for low latency and speed.
• The process can enrich a CRM with company mentions and generate market maps to support investment decisions.
• More videos from this channel:
  • How to Improve AI Coding With Better Context
  • How to Build an AI Driven Pipeline for Clients
• More from this channel: How I AI

Questions & Answers

Q: How does the Parakeet Podcast Processor handle audio to text conversion?

The processor downloads the podcast file, uses ffmpeg to convert the audio format, and then employs Parakeet to transcribe the audio into text. This sequence allows the pipeline to produce a textual representation of the spoken content suitable for downstream processing and summarization.

Q: What role does transcript cleaning play in the workflow?

Transcript cleaning is used to improve readability and extraction accuracy by normalizing and removing filler words. While the speaker noted that advances in large language models reduced the need for aggressive cleaning, initial steps helped with earlier named entity extraction before relying on the LLM for results.

Q: How are daily outputs stored and organized?

Daily outputs are stored in a local duct DB which records which podcasts were processed on a given day. This enables repeatable, day by day generation of summaries, quotes, and investment theses, and ensures outputs remain organized by date for easy retrieval.

Q: What kinds of outputs are produced from the transcripts?

The pipeline generates a daily summary document that includes host and guest details, a comprehensive summarization, key topics, and key themes. It also extracts noteworthy quotes and offers a set of actionable investment theses for venture capital use cases.

Q: How are investment theses generated from podcast content?

Investment theses are produced by prompts that extract actionable ideas and potential market opportunities discussed in the podcasts. The system uses the extracted quotes and topics to seed theses, which can then lead to market maps and targeted opportunities for research and outreach.

Q: Why is a terminal interface favored in this setup?

The creator prefers the terminal due to its low latency and speed, which reduces friction when interacting with complex AI workflows. This choice supports rapid scripting, batch actions, and quick feedback cycles essential for daily podcast processing.

Q: Can this workflow generate blog posts in the creator’s style?

Yes, the workflow includes prompts to draft blog posts in the writer’s voice. After transcribing and summarizing, the system can generate initial drafts that reflect the author’s style, which can then be iteratively refined to produce publishable content.

Q: What is the potential CRM integration mentioned in the workflow?

The system envisions enriching a CRM with mentions of companies discussed in podcasts, linking to notable startups like Airbnb or Google mentioned during episodes. This enables richer tracking of potential investments and relationships derived from podcast content.

Summary & Key Takeaways

• The approach centers on a daily pipeline that fetches, transcribes, and summarizes multiple podcasts to produce actionable insights.

• The workflow emphasizes structured outputs, including topics, themes, quotes, and potential investment ideas, making it easier to act on podcast content.

• The system demonstrates how AI and CLI tools can enable hyper personal productivity by generating drafts and prompts from podcast transcripts.