Ingestion

Ingestion is the point where you decide what ShrikeDB should know directly, what should stay in lakehouse files, and what meaning should live in profiles or ontologies.

Start simple: load a small slice, inspect it, validate the shape, then scale the same workflow.

Ingestion Choices

Source Use Best path

Small graph or app data

Direct graph/KV/vector writes.

gRPC or client API.

CSV/TSV tabular files

Load rows, infer columns, and create profile evidence.

Onboarding workflow or purpose-built loader.

RDF/Turtle/OWL

Load triples and ontology facts.

Ontology onboarding/import workflow.

Parquet

Inspect large data without copying it into graph storage.

Lakehouse source connection.

Existing semantic domain

Make natural-language questions reliable.

Profile pack import and validation.

A Good First Pass

inspect source
  -> load or register
  -> profile fields and relationships
  -> validate row counts and key columns
  -> import ontology/profile if needed
  -> run a few direct queries
  -> run semantic chat/MCP validation

Keep The Layers Separate

ShrikeDB works best when responsibilities stay clean:

  • raw records live in graph/KV/vector stores or external lakehouse files;

  • domain vocabulary lives in profiles;

  • ontology hierarchy lives in imported ontology facts;

  • reviewed lakehouse mappings connect external columns to graph concepts;

  • validation artifacts prove the workflow before you rely on the answers.

Related Pages