What is citymeetings.nyc? Who made it? How does it work?

Check out the about page to learn about the project.

Where do these meetings come from?

New York City Council meetings are downloaded from Legistar, which the city council uses to manage their meetings and legislative process.

The City Planning Commission's hearing on City of Yes for Housing Opportunity is from their channel on YouTube.

Can you cover other municipal or state government meetings?

I'd like to. I'm working on my software tools so I can do so scalably.

You can vote for which governing bodies you'd like me to tackle first by filling out this form.

Do you take feature requests?

Absolutely. Email me with your request.

Why are there errors in your transcripts? Won't they make your summaries inaccurate?

I use Deepgram's AI for transcription and diarization (a fancy word for "identifying when different speakers are speaking").

It's priced well for citymeetings.nyc but it's not perfect and it introduces two kinds of errors:

• Mistranscriptions, where the AI gets words wrong.
• Misdiarizations, where the AI gets speakers wrong.

You might see these errors when you view a transcript.

Large language models (LLMs) are very good at picking up on transcription and diarization errors using context clues.

I also employ techniques to make sure transcription errors don't creep into my summaries and reports.

Because it's not necessary to generate the artifacts you see on citymeetings.nyc, I don't manually review and fix all mistranscriptions and misdiarizations. It'd be too much work for my solo operation.

Occasionally things go awry. This happens with numbers and names of people/entities/locations in particular, and those numbers or entities might show up in the summary (e.g. someone said "70 million" but the transcript says "7 million").

Sometimes the errors are much worse, the mistranscription is offensive or entirely changes the meaning of what someone said.

I try to make sure these don't creep through, and am working on techniques to deal with them. If you see errors like the above, email me at vikram@citymeetings.nyc so I can fix them.

I have plans to try and use LLMs to fix transcription errors, but that is a large body of work and I'm prioritizing other things for now.

I see errors/inaccuracies. How do I report them?

Email me at vikram@citymeetings.nyc and read below for my policies, which I'm still figuring out and are subject to change.

My intent with summaries and reports is to faithfully represent things people have stated on the record in a way that is easy to skim.

If a chapter summary is inaccurate because it summarizes an inaccurate thing that something someone said, I won't correct the inaccuracy. It's on the record.

I don't do or show fact checks today. I might in the future.

I attribute claims made in summaries to speakers in meetings. These claims are not citymeetings.nyc's own claims. I'm working on making that clear in the summaries and the citymeetings.nyc UI.

If I have inaccurately represented something someone said, I'm very keen to fix it. This includes errors of omission. There's a tension between summary skimmability and how faithful they are to what someone actually said. I want to strike the right balance.

The most common errors you'll find are misspellings of names, entities, and locations because of transcription errors. I try to fix them before publishing meetings, am working on techniques to do so scalably, and will fix them if you email me.