I just read the second post in FollowFox’s series on using LLM pair programming to digest the Distillery codebase. The idea is how someone who doesn’t write code can use a big model to really wrap their head around a large project. The first post was about dumping the whole codebase into the model so it can answer questions, and this one mainly tackles a real-world problem: the model spits out way too much stuff, and chat interfaces and blogs just aren’t great for keeping it organized. They ended up going with Obsidian for storage and using the official Publish feature for sharing.
What I found most interesting was getting Claude to distill the main author’s coding style and put together a Style Guide. Even the author admitted their style is a bit non-pythonic, but once the model extracted those habits into a document, reading the code became way smoother for others. They also dug deep into the Discord Bot part, breaking down initialization, slash commands, the image generation flow, and queue processing in detail. The point wasn’t the final document itself, but the hands-on practice during the process. I totally vibe with that mindset.