Filogram

Context

An engineering organization accumulates decades of files across version-control servers and local drives. The knowledge is all there, and none of it is reachable.

Constraint

The corpus resists search on its own terms. Filenames are a mix of German and English, spelling and umlaut encoding are inconsistent, folder structures were invented per project, and duplication is heavy: revisions, backups, and verbatim copies of the same document. Searching such a corpus returns noise, and nobody knows what exists in the first place.

Contribution

I built Filogram, which turns that corpus into a navigable, clustered, faceted map. The primary lens is clustering and tagging, giving a thematic view of what exists where; fuzzy search and duplicate detection are supporting capabilities used to pinpoint individual files inside the map.

Approach

Normalization comes before analysis: umlaut folding and a curated German-to-English lexicon run first, so clustering and tagging operate on comparable text rather than on encoding accidents. The taxonomy and lexicon are configuration files treated as the runtime source of truth, so the vocabulary can be corrected without touching the code.

Outcome

The question changes from find this file, which assumed you knew it existed, to show me what is here, which is the question people actually have.

Evidence

The work is held in a private repository, with the domain glossary, the data characteristics, and the live versus dormant capabilities documented explicitly.