Cancelling Projects

I finally did it, instead of spending way too much time on a simple side-project I’ve been thinking and tinkering about, I let go of it.

Way too many times software developers end up in starting more and more (side-)projects while never either finishing or cancelling old ones. They become abandoned while still taking up head-space somewhere in a dark corner of the mind. This post will be a list of my abandoned or cancelled projects, including short descriptions of what I intended to achieve and where i think I failed.

Prollama

Prollama was my attempt to create an automated Pull Request Review Bot for Github. The clue: it’s supposed to be a) zeroconf (you need gh and ollama configured, though) and privacy first. Your code/review-data will never leave your own computer, there’s no Remote API involved (if you don’t want to). The idea was simple enough to have an MVP ready in an evenings’ work. The path to stability, though, was - as it is often - totally underestimated by me. So after trying to figure out how to extract hunks from a unified diff file I had to prompt engineer my way through different local models to figure out how to get a model to respond in a structured format. And i had to unescape/unquote/uncontrolcharacter the returned string, which sometimes turned into a mess because suddenly things like ‘+’ caused the unmarshaller to escape the next character (e.g., the i of import slog). This in turn then caused the unmarshaller to return an error and rendered the whole hunk unusable. And this was only the beginning, so I pulled a canary and escaped as soon as I figured this would not end up well for me. Maybe someday this project will be activated again, but until then everyone is free to learn from it or even continue the work on it.

But as always, you learn from those projects, and this is what I’ve learned when working on Prollama: