Live data from Hacker News

Show HN: A Tool to Summarize Kenya's Parliament with Rust, Whisper, and LLMs

github.com

11–14 of 14 posts

Re: Show HN: A Tool to Summarize Kenya's Parliament with Rust, Whisper, and LLMs

#12
post #10

If you wanted to make the transcription locally hosted to save on OpenAI costs, you can use my crate, mutter [0], which makes self-hosting the Whisper model super easy :) [0]: https://crates.io/crates/mutter

Can I box this up in a docker container and stuff it behind an API?

Absolutely! There’s existing solutions for this, if you hunt around, but if you wanted to customize it, it would be pretty easy.

Re: Show HN: A Tool to Summarize Kenya's Parliament with Rust, Whisper, and LLMs

#13

Nice! Love to see initiatives like these. I've been working on something in the same space for the Belgian federal parliament. The Belgian parliament livestreams sessions and publishes a single (long, bloated, dual-language) PDF report[0] for each session and that's it. This means no search across sessions, no details of which parties voted how, no API etc. The only view you get is from the perspective of a single se…

Any chance of this being open-source? I would like to build something like this for my country as well.

Re: Show HN: A Tool to Summarize Kenya's Parliament with Rust, Whisper, and LLMs

#14

Nice! Love to see initiatives like these. I've been working on something in the same space for the Belgian federal parliament. The Belgian parliament livestreams sessions and publishes a single (long, bloated, dual-language) PDF report[0] for each session and that's it. This means no search across sessions, no details of which parties voted how, no API etc. The only view you get is from the perspective of a single se…

Any chance of this being open-source? I would like to build something like this for my country as well.

Not at the moment but maybe in the future. While I'm sure the code could be of interest, the code definitely isn't a template that would fit for other countries. It is tightly coupled to the format of the Belgian parliament since most of it is parsing their PDF reports. In detail my tech stack is like this:

- Rust + reqwest crate + scraper crate to download reports and parse data from it into parquet files

- 11ty + @duckdb/node-api to parse the parquet files at build time and generating pages automatically, I have for example a 'meeting' template and then for each meeting a html page gets built

- summarize certain data using mistral API

- deploy on cloudflare free plan

Post reply on HN