EvoOntology: a self-evolving ontology layer bridges the agent-data gap

RUC-DataLab · hf · 2026-09-21

RUC-DataLab introduces EvoOntology, a self-evolving ontology layer for data agents struggling with heterogeneous data (tables, files, databases).

Across three data-agent benchmarks and four LLM backbones, EvoOntology consistently beats strong baselines and prior semantic-layer approaches. Code is open source at github.com/ruc-datalab/EvoOntology.

Original post →

More from coding & agent

coding & agent channel →