fvolkek
Master
- Joined
- Feb 19, 2021
- Posts
- 1,521
- Reputation
- 1,805
Not really work but it's hard to explain in the title. Tomorrow I have a sort of "MUN"-ish competition, but instead of MUN, it's my country's politics. Hard to explain, but it's university level and a big deal. And I'll represent my university in the parlament, and I want to optimize my workflow. Since I'm also a AI Agent freak, I might as well make a workflow with agents.
The simulation is this, we built a draft project (mine will likely be voted), and then we will amend article by article, and discuss it. That's it. Lobbying and all of that will be done, but it's trivial. I want to have several agents with different tasks. I have a lot of compute so don't worry.
The idea is to have all the Argentine legal documents (especially regarding gambling, which is what the debate is about), so the constutiton, criminal code, civil code, commercial code, and whatnot. Jurisprudence too, I guess. And for it to know when to route to each, so when to consult each not to waste the context window. All of them being in .md to be token efficient.
I will obviously have the creative and final decision, I want there to always be a human in the loop, but the idea is to have some dude who tells me "wait, what he's saying violates art. 1231 of the criminal code and art. 33 of criminal code, and the consitution too, and jurisprudence says x", whatever. You get the gist. Apologies for my English, too.
Any ideas?
The simulation is this, we built a draft project (mine will likely be voted), and then we will amend article by article, and discuss it. That's it. Lobbying and all of that will be done, but it's trivial. I want to have several agents with different tasks. I have a lot of compute so don't worry.
The idea is to have all the Argentine legal documents (especially regarding gambling, which is what the debate is about), so the constutiton, criminal code, civil code, commercial code, and whatnot. Jurisprudence too, I guess. And for it to know when to route to each, so when to consult each not to waste the context window. All of them being in .md to be token efficient.
I will obviously have the creative and final decision, I want there to always be a human in the loop, but the idea is to have some dude who tells me "wait, what he's saying violates art. 1231 of the criminal code and art. 33 of criminal code, and the consitution too, and jurisprudence says x", whatever. You get the gist. Apologies for my English, too.
Any ideas?
