From DataFrame to Knowledge Graph in 4 lines of Python w/ Veronika Heimsbakk

TL;DR

• 1

  Four lines gets you a working graph: Veronika shows Polars + maplib + Model + map_default turning a CSV into a knowledge graph immediately, then jokes that the talk could end there because the basic promise is already fulfilled.

• 2

  The key idea is triples plus global IDs: Every row becomes explicit subject-predicate-object triples, and shared identifiers like band names let data from bands, vinyl inventory, and festival lineups merge into one graph.

• 3

  SPARQL looks familiar if you know SQL: Veronika frames SPARQL as SQL-like but graph-native, with no explicit joins because the graph's identifiers already define the connections, and SELECT queries return dataframes in maplib.

• 4

  Strings become things with meaning: She upgrades raw genre strings into SKOS concepts, so 'blackgaze' is no longer just text but a concept with broader and narrower relationships, preferred labels, and alternate labels.

• 5

  The demo stays practical, not academic: The graph combines 16 bands, a fictional cheap record store, and fictional festival lineups into concrete queries like 'which Roadburn bands have albums in stock and at what price?'

• 6

  Scale was a direct audience concern: Asked how this holds up in production, Veronika says RDF scaling problems are old news, and that maplib can handle up to about 3 billion triples on disk before breaking on a regular Mac.