# Data Ingestion

This page summarizes the ingestion approach used in the demo repository, focusing on reusable patterns rather than dataset specifics.

1. Define node and edge schemas (or validate they exist).
2. Upsert nodes by stable keys.
3. Create relationships via edges.
4. Commit in batches.

The demo connector illustrates this structure:

await graph.CreateNodeSchemaAsync<Schema.Nodes.Device>();
await graph.CreateNodeSchemaAsync<Schema.Nodes.Part>();
await graph.CreateEdgeSchemaAsync(typeof(Schema.Edges));

var deviceNode = graph.TryAdd(new Schema.Nodes.Device { Name = device.Name });
var partNode   = graph.TryAdd(new Schema.Nodes.Part   { Name = part.Name });
graph.Link(partNode, Node.FromKey(nameof(Schema.Nodes.Device), device.Name),
           Schema.Edges.PartOf, Schema.Edges.HasPart);

await graph.CommitPendingAsync();

• Use stable keys to keep ingestion idempotent.
• Encode relationships explicitly (don’t expect them to be inferred).
• Add observability (counts, durations, error logs).
• Plan for schema evolution and reprocessing.

• Connector guidance: Connectors
• Pipeline guidance: Ingestion Pipelines