Conduit Podcast Transcripts

This is an archive of transcriptions generated by NVIDIA Parakeet, Audio Hijack, and related tools meant to be used as a source of data.

The text is from the Conduit Podcast

Getting Started

Installation

This project uses uv for fast, reliable Python package management. To set up:

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies and create virtual environment
uv sync

Environment Configuration

Load the environment variables:

# Using direnv (recommended)
direnv allow

# Or manually source the .envrc file
source .envrc

The .envrc file should contain connection strings for PostgreSQL and other configuration.

Usage

The project provides multiple interfaces for accessing and managing transcripts:

• CLI Tool - conduit command for local operations
• REST API - FastAPI server for programmatic access
• MCP Server - Model Context Protocol server for Claude integration

Docker Setup

Start all services with Docker Compose:

# Start PostgreSQL and API server
docker compose up -d

# View logs
docker compose logs -f

# Stop services
docker compose down

The API will be available at http://localhost:8000 with interactive docs at /docs.

CLI Usage (via Docker)

docker compose run --rm app python -m cli.main [command] [options]

Transcription

Transcribe episodes from the Conduit website using NVIDIA Parakeet:

# Transcribe the latest episode
docker compose run --rm app python -m cli.main transcribe <episode_number>

# Transcribe and ingest (default)
docker compose run --rm app python -m cli.main transcribe <episode_number>

# Use specific model size/name (default: nvidia/parakeet-rnnt-1.1b)
docker compose run --rm app python -m cli.main transcribe <episode_number> --model nvidia/parakeet-rnnt-1.1b

# Configure the LLM model for RAG (Retrieval Augmented Generation)
docker compose run --rm app -e LLM_MODEL=llama3 python -m cli.main transcribe <episode_number>

Data Ingestion

Load transcripts into PostgreSQL:

# Ingest all files in transcripts directory
docker compose run --rm app python -m cli.main ingest

# Ingest specific file
docker compose run --rm app python -m cli.main ingest --file transcripts/episode1.md

# Recreate tables before ingestion
docker compose run --rm app python -m cli.main ingest --reindex

Search

Search through ingested transcripts:

# Text search (default)
docker compose run --rm app python -m cli.main search "search term"

# Vector semantic search
docker compose run --rm app python -m cli.main search "search phrase" --vector

MCP Server Usage

The project includes a Model Context Protocol (MCP) server that allows AI assistants (like Claude) to directly query the transcript database.

Configuration

Add the server to your MCP client configuration (e.g., claude_desktop_config.json):

{
  "mcpServers": {
    "conduit": {
      "command": "docker",
      "args": [
        "compose",
        "run",
        "--rm",
        "app",
        "python",
        "-m",
        "app.mcp.server"
      ]
    }
  }
}

Or if connecting to a running instance (e.g. via SSE):

{
  "mcpServers": {
    "conduit": {
        "url": "https://conduit.kjaymiller.dev/mcp/sse",
        "transport": "sse"
    }
  }
}

Available Tools

The MCP server provides the following tools:

• search_transcripts: Search through transcripts using keyword or vector search.

  • query (string): The search text.
  • limit (int, optional): Max results (default 10).
  • use_vector (bool, optional): Use semantic vector search (default True).
  • episode_number (int, optional): Filter by episode number.

• get_episode: Retrieve full content and metadata for an episode.

  • episode_number (int): The episode number to retrieve.

• list_episodes: List available episodes with metadata.

  • limit (int, optional): Max results (default 20).
  • start_date (string, optional): Filter by start date (YYYY-MM-DD).
  • end_date (string, optional): Filter by end date (YYYY-MM-DD).

Management

# Check episode status
docker compose run --rm app python -m cli.main status <episode_number>

# List recent episodes
docker compose run --rm app python -m cli.main list

Project Structure

• app/ - FastAPI application
  • api/ - REST API endpoints (search, episodes, health)
  • mcp/ - MCP Server for Claude integration
  • main.py - Main FastAPI application
• cli/ - Command-line interface
• podcast_transcription/ - Shared library code
  • database/ - Database operations (PostgreSQL)
  • models/ - SQLAlchemy models
  • transcription/ - Transcription logic (NVIDIA Parakeet)
  • utils/ - Shared utilities
• transcripts/ - Generated markdown files with metadata and transcriptions

Troubleshooting

Virtual environment issues: Run uv sync and ensure you're using uv run or have activated the venv

Missing environment variables: Load .envrc with direnv allow or source .envrc

Parakeet model: First run downloads the default model (nvidia/parakeet-rnnt-1.1b) - requires network access

Database connection: Services are on Aiven; verify credentials and network access

Technology Stack

• Python 3.13+
• NVIDIA Parakeet (transcription)
• LangChain (text processing)
• PostgreSQL with pgvector extension
• SQLAlchemy (ORM)
• Click (CLI framework)
• FastAPI (REST API)
• MCP (Model Context Protocol)

Usage and License

Conduit Podcast Transcripts by Jay Miller, Kathy Campbell, original downloads from whisper work done by Pilix is licensed under Attribution-NonCommercial-ShareAlike 4.0 International