Extract: Turning Unstructured Data into Structured Entities

Raw emails and documents are unstructured, free-form text, with inconsistent formats, and implicit relationships. The extraction layer transforms this content into typed, structured entities that the graph can reason on.

Our proprietary AI extraction engine classifies content for relevance, parses text from attachments and performs structured extraction into validated entity schemas. Every extracted entity is schema-validated before entering the graph, ensuring no untyped or malformed data gets through.

What Gets Extracted

Extraction is source-agnostic. Whether an HBL number appears in a subject line, an email body, or a scanned PDF, it enters the graph as the same entity type. The same identifier appearing across multiple sources strengthens the graph rather than creating duplicates.

Every piece of content can produce a rich set of logistics entities:

Identifiers: The reference numbers that tie logistics operations together. These are the most critical extraction, as they're how shipments are tracked across systems. For example, below are of the identifiers we look at

Type	Example	Purpose
House Bill of Lading	TPEB1234567	Primary shipment reference
Booking Number	BKG-2024-0892	Carrier booking reference
Master Bill of Lading	MAEU123456789	Consolidation-level reference
Purchase Order	PO-44210	Commercial reference
Invoice Number	INV-2024-3391	Financial reference
Container Number	MRKU4512870	Physical unit tracking

Identifiers also carry relationships to each other — an invoice may reference a booking, a master bill may contain multiple house bills:

Parties & Contacts: Organizations (shipper, consignee, carrier, forwarder, customs broker) and the people within them, with their specific roles on each shipment.

Route & Cargo: Multi-modal route legs (ocean, truck, rail, air) with origin/destination locations, vessel/flight info, and ETD/ETA timestamps. Containers with type and seal info. Commodities with HS codes, weight, and dimensions.

Financial: Charges with amount, currency, and billed-by/billed-to parties. Rate cards and quotations linking pricing to service providers.

Events: Milestone events (booking confirmed, cargo received, vessel departed, customs cleared, delivered) with timestamps and status tracking (estimated, planned, actual).