GeoCompress

GeoCompress is a compression system that treats data as geometric objects and uses signature analysis to discover optimal compression transform chains through evolutionary search. The system is implemented in Rust across four crates totaling roughly 353,000 lines.

The approach is deliberately unlike standard compression pipelines. Instead of applying one fixed stack of transforms to everything, GeoCompress tries to characterize the input first and then search for a transform chain that fits that particular shape.

Why it is different

Most familiar compression tools embody one strong opinion about what data looks like and then apply that opinion broadly. GeoCompress treats that as a limitation.

The core bet is that data families have geometric signatures that can be detected and exploited. If that signature can be measured usefully, the compressor does not need to settle for a one-size-fits-all pipeline. It can search for a chain of transforms that is better matched to the structure of the input in front of it.

What the project actually is

This is not a toy experiment or a one-file algorithm sketch. The current system spans four Rust crates and roughly 353,000 lines.

That scale exists because the project is doing more than a single clever encoding trick. It has to represent transforms, compare candidate chains, and make the search process concrete enough to run as a real system instead of as a hand-wavy idea.

Current posture

GeoCompress is still work in progress, but the interesting thing about it is already clear: it treats compression as a search problem over structure rather than as a fixed recipe.

That makes it one of the more experimental projects on the site, but also one of the clearest examples of the kind of systems work I like most: pick a strong underlying model, take it seriously, and build until it either proves itself or fails honestly.