Concepts and relations behind every dataset
A live view of the portal's knowledge graph — what we know about each dataset and how it connects to the rest.
By entity type
Top relations
By dataset
1 · Authors annotate MDX
Each dataset entry tags concepts inline: {{entity:concept:lane-detection}} and {{relation:supports_task:bdd100k:lane-detection}}.
2 · Sync extracts to Postgres
A CI script writes nodes and edges into knowledge_graph_nodes and knowledge_graph_edges, scoped by dataset:<slug>.
3 · Worker embeds each node
A pgvector column on every row holds a 1536-d Cohere multilingual v3 embedding for semantic recall.
4 · Cross-dataset unification
Same slug across datasets resolves to one Global concept via an application-layer join — local KGs merge into a Global KG.
Example annotation from a dataset's MDX entry
BDD100K supports {{entity:concept:lane-detection}} as a primary task.
It was published by {{entity:organization:berkeley-deepdrive}}.
{{relation:supports_task:bdd100k:lane-detection}}
{{relation:published_by:bdd100k:berkeley-deepdrive}}