This project takes a YouTube podcast URL, extracts the transcript, identifies key topics and Q&A pairs, simplifies them for children, and generates an HTML report with the results.

1. LLM Calls (utils/call_llm.py)

2. YouTube Processing (utils/youtube_processor.py)

   • Get video title, transcript and thumbnail

3. HTML Generator (utils/html_generator.py)

   • Create formatted report with topics, Q&As and simple explanations

The application flow consists of several key steps organized in a directed graph:

1. Video Processing: Extract transcript and metadata from YouTube URL
2. Topic Extraction: Identify the most interesting topics (max 5)
3. Question Generation: For each topic, generate interesting questions (3 per topic)
4. Topic Processing: Batch process each topic to:
   • Rephrase the topic title for clarity
   • Rephrase the questions
   • Generate ELI5 answers
5. HTML Generation: Create final HTML output

flowchart TD
    videoProcess[Process YouTube URL] --> topicsQuestions[Extract Topics & Questions]
    topicsQuestions --> contentBatch[Content Processing]
    contentBatch --> htmlGen[Generate HTML]
    
    subgraph contentBatch[Content Processing]
        topicProcess[Process Topic]
    end

The shared memory structure will be organized as follows:

shared = {
    "video_info": {
        "url": str,            # YouTube URL
        "title": str,          # Video title
        "transcript": str,     # Full transcript
        "thumbnail_url": str,  # Thumbnail image URL
        "video_id": str        # YouTube video ID
    },
    "topics": [
        {
            "title": str,              # Original topic title
            "rephrased_title": str,    # Clarified topic title
            "questions": [
                {
                    "original": str,      # Original question
                    "rephrased": str,     # Clarified question
                    "answer": str         # ELI5 answer
                },
                # ... more questions
            ]
        },
        # ... more topics
    ],
    "html_output": str  # Final HTML content
}

• Purpose: Process YouTube URL to extract video information
• Design: Regular Node (no batch/async)
• Data Access:
  • Read: URL from shared store
  • Write: Video information to shared store

• Purpose: Extract interesting topics from transcript and generate questions for each topic
• Design: Regular Node (no batch/async)
• Data Access:
  • Read: Transcript from shared store
  • Write: Topics with questions to shared store
• Implementation Details:
  • First extracts up to 5 interesting topics from the transcript
  • For each topic, immediately generates 3 relevant questions
  • Returns a combined structure with topics and their associated questions

• Purpose: Batch process each topic for rephrasing and answering
• Design: BatchNode (process each topic)
• Data Access:
  • Read: Topics and questions from shared store
  • Write: Rephrased content and answers to shared store

• Purpose: Create final HTML output
• Design: Regular Node (no batch/async)
• Data Access:
  • Read: Processed content from shared store
  • Write: HTML output to shared store