have done similar project (youtube summarization and question answeirng).

I find that youtube video meta infor, especially video title and descriptions are quite helpful for LLM to better understand the context.

and similarly, thumbnail image url can be useful when build web app with this.

a sample output would be
{'source': 'I845O57ZSy4',
 'title': 'John Carmack: Doom, Quake, VR, AGI, Programming, Video Games, and Rockets | Lex Fridman Podcast langchain-ai#309',
 'description': 'John Carmack is a legendary programmer, co-founder of id Software, and lead programmer of many revol...',
 'view_count': 1229347,
 'thumbnail_url': 'https://i.ytimg.com/vi/I845O57ZSy4/sddefault.jpg',
 'publish_date': datetime.datetime(2022, 8, 4, 0, 0),
 'length': 18890,
 'author': 'Lex Fridman',
}

Love the work you guys have done on LangChain, hope i can also contribute a tiny bit!

…ntext will be retrieved

given that we allow user define chunk size, think it would be useful for user to define how many chunks of context will be retrieved. or the default 4 chunk can break some code.