Does It For You

A clever AI construction kit

Somewhere, out there, in Signal chat room land, there’s an AI related group. There are two other guys a bit like me - they own their own cloud computing setups. Their practices are a bit like mine - knowledge engineering stuff. There are three people who do not have any formal programming experience, but they’re tinkering with Python and AI related tools. The other half of the room are those who would benefit from getting some AI stuff implemented, they’re shot callers in their respective organizations.

About two months ago I first encountered Dify.ai and tonight, thanks to Matt Williams, I now know it stands for Does It For You. If you’ve ever used n8n for automation, or Orange Data Mining, you’ll find Dify’s construction style immediately familiar.

September has begun and my technology queue has cleared, so this is the month I’m going to get SOMETHING running within Disinfodrome that uses Dify to assist users in finding things. I’m following this tutorial to get started.

There are ways to do this on a bare Linux distro but I’m also biting the bullet and shifting all of this stuff to Docker. This is what I see running as I’m nearing the end of the tutorial video.

There’s another non-AI thing that I’ve got going as well. Disinfodromes’s front end has been an Nginx verse proxy since its inception. Last month, while tinkering with MediaWiki, I encountered the Traefik reverse proxy for the first time. It’s got a fearsome learning curve but it looks like something an enterprise oriented guy like me might be using. I found this video tutorial that compares Traefik with Nginx Proxy Manager. I’ve always just whacked Nginx into operation the old fashion way, using a text editor and systemctl. I’d never considered that there might be a handy web front end for doing that.

And there’s a silent AI partner in all this - ChatGPT. It can do a good job with awk, it’ll solve complex rsync problems without much trouble, and if you’re picky about what you ask you can get it to produce workable Python code as well. I’ve been describing it as my intern and I think that’s an apt description - I give it tasks that are conceptually simple, but which would take me an hour with Google to find something close, and it coughs up an answer in a few seconds.

So ChatGPT is productive for me, but there isn’t anything in Disinfodrome that is productive for someone else because it’s AI enabled. That needs to change this month.