Podcast fetcher

AI Podcast Manager Telegram bot that finds, transcribes & summarizes podcasts. Instant bullet-point summaries of your favorite shows delivered to your phone - either immediately, daily or weekly.

Comment

Podcast Fetcher

[I couldn't upload the video to a different platform, please consider the video below as my demo video and ignore the (invalid) Youtube one I entered as video demo link] https://www.loom.com/share/8f33e501a5794f84984035c6ff96637e?sid=f9b4b361-1225-4bfd-8ed7-85f92f86caff

About the Project

Podcast Fetcher is an AI-powered Telegram bot that improves the way people consume podcast content. The project was born from a personal frustration: it was impossible to keep up with the ever-growing number of AI-focused podcasts, much less to consume all of them and keep notes. The bot intelligently searches for podcasts using the Taddy API, automatically transcribes audio using AWS Transcribe, and generates concise AI-powered summaries using AWS Bedrock. Users receive personalized bullet-point summaries of their subscribed podcasts directly in Telegram, transforming hours of browsing into seconds of reading.

What Inspired Us

The inspiration came from the explosion of podcast content during the pandemic and the realization that most people have limited time to consume long-form audio content. We wanted to create a solution that would:

  • Make podcast discovery effortless through intelligent search
  • Provide instant value through AI-generated summaries
  • Deliver content in the most accessible format (mobile messaging)
  • Scale automatically to handle thousands of users and podcasts

How We Built It

Architecture & Technology Stack

  • Frontend: Telegram Bot API with conversation flows and inline keyboards
  • Backend: Python with FastAPI, deployed on Modal for serverless scalability
  • AI/ML Pipeline: AWS Transcribe for speech-to-text, AWS Bedrock for summarization
  • Database: Supabase
  • Search: Taddy API integration for podcast discovery
  • Deployment: Modal for serverless functions with persistent storage

Key Components

  • Smart Search System: Integrated Taddy API to search podcasts by name/topic with structured data models
  • Automated Processing Pipeline: RSS parsing → Audio download → Transcription → AI analysis → User delivery
  • Conversation Management: Intuitive Telegram bot with subscription flows and user preferences
  • AI Summarization: Custom prompts for 3-bullet point summaries using AWS Bedrock
  • User Management: Subscription tracking, notification preferences, and episode lookback settings

Technical Challenges Solved

  • Audio Processing: Handled various audio formats and large file downloads using AWS Transcribe
  • Scalability: Designed serverless architecture that can handle thousands of concurrent users

Challenges We Faced

  • Conversation Flow Design: Creating intuitive user experiences in a text-based interface was challenging
  • Deployment Complexity: Setting up Modal with persistent storage and proper secret management

Accomplishments We're Proud Of

  • Complete End-to-End Pipeline: Built a fully automated system from podcast discovery to user delivery
  • AI Integration: Successfully integrated multiple AI services for transcription and summarization
  • User Experience: Created an intuitive Telegram bot with subscription management and preferences
  • API Integration: Successfully integrated three different APIs (Taddy, AWS, Telegram) with robust error handling

What We Learned

  • Database Modeling: Designed efficient schemas for complex user-podcast relationships
  • Conversation Design: Mastered the art of creating intuitive text-based user interfaces

What's Next

  • Handle multiple audio languages - better leveraging aws transcribe
  • Use speaker diarization (consider pyannote)
  • Be a proactive agent, i.e. actively suggest podcast episodes
  • Include memory into bot - act like a personal note taker

Built With

Share this project:

Updates