🎙️ Podcast AI Assistant & Content Generator

An intelligent application that transforms long-form podcast transcripts into a queryable knowledge base. Ask questions, get summaries, and generate new content in seconds. This project demonstrates an end-to-end Retrieval-Augmented Generation (RAG) pipeline, from data processing to a user-friendly web interface.

---

✨ Features

• Conversational Q&A: Ask questions in natural language and get concise, context-aware answers sourced directly from the podcast transcripts.
• On-the-Fly Summarization: Generate summaries of specific topics or entire episodes.
• Content Extraction: Pull out key takeaways, mentioned names, or companies in a structured format.
• Content Repurposing: Transform podcast discussions into new formats like Twitter threads or blog post outlines.
• Source Citing: Each response includes the specific text chunks from the transcripts that were used to generate the answer, ensuring transparency and trust.

---

🛠️ Tech Stack & Architecture

This project is built on a modern, modular stack designed for building LLM applications.

• LLM Framework: LlamaIndex for orchestrating the RAG pipeline.
• LLM & Embeddings: Google Gemini API (gemini-1.5-flash-latest for generation, embedding-001 for embeddings).
• Web Framework: Streamlit for creating the interactive user interface.
• Language: Python 3.10+
• Environment: python-dotenv for managing API keys.

Architecture Overview

The application follows a classic RAG pattern:

1. Offline Indexing (engine.py):

   • Podcast transcripts are loaded from the /data directory.
   • Text is split into smaller, manageable chunks.
   • Each chunk is converted into a vector embedding using Gemini's embedding model.
   • The embeddings and corresponding text are stored in a local vector index using LlamaIndex's StorageContext.

2. Real-time Querying (app.py):

   • A user submits a query through the Streamlit interface.
   • The query is converted into an embedding using the same model.
   • The vector index is searched to find the most semantically similar text chunks (the "context").
   • The context and the original query are passed to the Gemini Pro model in a carefully crafted prompt.
   • The LLM generates a response based only on the provided context, which is then streamed back to the user.

---

🚀 Getting Started

Follow these instructions to set up and run the project locally.

Prerequisites

• Python 3.10 or higher
• A Google Gemini API Key

1. Clone the Repository

git clone https://github.com/your-username/podcast-qa.git
cd podcast-qa

2. Set Up a Virtual Environment

Windows:

python -m venv .venv
.\.venv\Scripts\activate

macOS / Linux:

python3 -m venv .venv
source .venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure Your API Key

1. Create a file named .env in the root of the project directory.

2. Add your Gemini API key to this file:

   GEMINI_API_KEY="YOUR_API_KEY_HERE"

5. Prepare Your Data

• Place your podcast transcript .txt files inside the /data directory.

6. Build the Vector Index

Run the indexing script. This only needs to be done once, or whenever you add new documents to the /data folder.

python engine.py

This will create a storage directory containing your vector index.

7. Run the Streamlit App

You're all set! Launch the application with:

streamlit run app.py

Your web browser should automatically open with the app running.

---

💡 Future Improvements

• Implement conversation memory to allow for follow-up questions.
• Add support for more data sources (e.g., YouTube video transcripts, RSS feeds).
• Explore more advanced retrieval strategies like hybrid search.
• Containerize the application with Docker for easier deployment.

---

📜 License

This project is licensed under the MIT License. See the LICENSE file for details.