fetch event recordings from youtube

[david-wolgemuth]

• add model EventRecording
• fetches Matt's youtube playlists and relates them to groups

misc

• adds package https://github.com/django-extensions/django-extensions (for shell_plus)

todos / followups

• build a frontend for viewing these

• relate recordings to events ; for now only relates to groups. will probably need to manually relate to events?

  • (could probably use AI to quick do this ; but don't want to muddy up this pr)
  • only 5 events match the recording titles as is:

>>> rs = EventRecording.objects.all()
>>> for r in rs:
...     if Event.objects.filter(name=r.title).exists():
...         print(r.title)
...
Boosting AI with Python: Using Click, Jinja2, and GPT Libraries
NATS: Connecting Apps Over a Network Easily
An Opinionated Introduction to CI/CD
NATS: Connecting Apps Over a Network Easily
No Frills, Just Go: Standard Library Only Web Apps

Visualization

[mblayman]

Looks cool.

What would be the plan for fetching these and knowing how/when to link them? I'm not eager to sign up for manual work.

[mblayman]

I don't reliably set the speaker in a consistent way, so I agree that we should probably get speaker data from a different source.

[david-wolgemuth]

What would be the plan for fetching these and knowing how/when to link them? I'm not eager to sign up for manual work.

Yeah, I wasn't sure if you had thoughts on that

A couple ideas I had:

• when you upload recording, include the related meetup URL in the description
• AI script - I just pushed up an example of what that might look like (i can flush it out / clean it up if you like that idea)

[david-wolgemuth]

example run of this script,

$ uv run manage.py resolve_event_recording_names

Recording: recording.title='PDF Text Extraction With Python' recording.published_at=datetime.datetime(2024, 8, 15, 3, 53, 12, tzinfo=datetime.timezone.utc)
Event: event.name='PDF Text Extraction (2nd Wed Talk)' event.start_at=datetime.datetime(2024, 8, 14, 23, 0, tzinfo=datetime.timezone.utc)

(i'd clean this up / complete it if you like the idea)

[mblayman]

This seems cool. I guess we'd only need to run this when there is a recording that doesn't have an event attached, so the actual API calls would be super low. Also, if we are pulling from the known playlists where I'd be adding recordings, there could be a pretty high degree of confidence in the matching.

You've probably done more OpenAI stuff than I have. How deterministic do you think this is likely to be?