• Workflow
  • User input a prompt about what news they want to know about.
  • Each day the agent will search the web for news that fix the prompt that user set.
  • The agent will classify the news into categories based on the relevance to the prompt.
  • It will then generate an audio as a form of podcast for the listener.
• analyze (prompting.py):
  • Input: User prompt
  • Output: list[interest] List of user interest
• crawl (crawl.py):
  • Input: news website, # of article to crawl
  • Output: list[article] List of article object
• gen_embed (crawl.py):
  • Input: list of article
  • Update embedding library
  • Output: Dict{article, embedding}
• rank (crawl.py):
  • Input: Dict{article, embedding}, list[interest]
  • Output: Dict{interest, list[article]} List of article according to interest
• generate_podcast_script (generation.py):
  • Input: Dict{interest, list[article]}
  • Output: list[Line] List of script lines
• synthesize_audio (generation.py):
  • Input: list[Line]
  • Output: mp3

V 0.2

• Crawl Pipeline:
  • crawl (crawl.py):
    • Input: news website, # of article to crawl
    • Output: list[title, url, summary]
  • gen_embed (crawl.py):
    • Input: list[title, url, summary]
    • Update embedding library (.pkl): Dict{url, embedding}
    • Output: None
• Recommendation Pipeline
  • analyze (prompting.py):
    • Input: User prompt
    • Output: list[interest] List of user interest
  • rank (crawl.py):
    • Input: Read .pkl file: Dict{url, embedding}, list[interest]
    • Download article based on url
    • Output: Dict{interest, list[article]} List of article according to interest
  • generate_podcast_script (generation.py):
    • Input: Dict{interest, list[article]}
    • Output: list[Line] List of script lines
  • synthesize_audio (generation.py):
    • Input: list[Line]
    • Output: mp3