Extraction Profiles

DocIntel uses domain-aware extraction profiles — each with its own ontology (entity types + relationship types), chunking strategy, model tier, and confidence threshold. Documents are auto-classified to the best profile, or you can specify one explicitly.

Profile summary

Profile	profile= value	Model tier	Chunking	Confidence threshold
AEC / BIM	aec	FLAGSHIP	table_aware	≥ 0.75
Legal / Contract	contract	FLAGSHIP	clause	≥ 0.75
Financial / ERP	financial	DEFAULT	table_aware	≥ 0.68
Scientific	scientific	DEFAULT	section	≥ 0.70
General	general	DEFAULT	paragraph	≥ 0.65

Auto-classification

When no profile parameter is provided, DocIntel classifies the document automatically:

1. Keyword heuristic fast path — scans the first 3,000 characters for domain-specific keywords (13–14 per domain). If ≥ 3 keywords match, the profile is selected immediately — no LLM call required.
2. LLM fallback — if the heuristic is inconclusive, a one-shot LLM classification determines the domain.

AEC / BIM

Use case: IFC models, CAD drawings, AEC specification documents

Entity types	Relationship types
IfcElement, IfcSpace, IfcMaterial, IfcSystem, RegulatoryCode, IfcClassification	MADE_OF, LOCATED_IN, CONTAINS, CONNECTS_TO, CLASSIFIED_AS, COMPLIES_WITH

The AEC profile uses the FLAGSHIP model tier (gpt-5.6-sol) and table-aware chunking to handle room schedules, material take-offs, and BIM property sets.

AEC use cases →

Legal / Contract

Use case: Contracts, NDAs, service agreements, procurement documents

Entity types	Relationship types
Party, Clause, Obligation, Right, MonetaryValue, Date	SIGNED_BY, OBLIGATES, GRANTS_RIGHT_TO, GOVERNS, REFERENCES

Uses clause-based chunking to respect section boundaries. The FLAGSHIP model and ≥ 0.75 threshold minimise hallucinated party names.

Legal use cases →

Financial / ERP

Use case: Invoices, purchase orders, financial statements, SAP IDocs

Entity types	Relationship types
Company, Transaction, GLAccount, Amount, InvoiceLineItem, Date	ISSUED_BY, RECEIVED_BY, POSTS_TO, REFERENCES, PAID_BY

Uses table-aware chunking to handle line items and financial tables. SAP IDoc segment structure (INVOIC02 / ORDERS05) is parsed automatically.

Finance use cases →

Scientific

Use case: Academic papers, technical reports, research articles

Entity types	Relationship types
Author, Institution, Dataset, MLModel, Finding, Method, Metric	AUTHORED_BY, AFFILIATED_WITH, TRAINED_ON, EVALUATED_ON, OUTPERFORMS, REFERENCES

Uses section-based chunking (abstract, methods, results, references) to respect paper structure.

Scientific use cases →

General

Use case: Enterprise knowledge, mixed-domain documents, catch-all

Entity types	Relationship types
Person, Organisation, Location, Product, Event, Policy, Concept, Date, Value, Reference, Document	HAS_DATE, HAS_VALUE, IDENTIFIED_BY, REFERENCES, RELATES_TO, PART_OF

The general profile uses paragraph-based chunking and the lowest confidence threshold (≥ 0.65) to maximise recall across diverse document types.

Enterprise use cases →

Forcing a profile

curl -X POST http://localhost:8200/process \
  -H "Authorization: Bearer $API_KEY" \
  -F "file=@document.pdf" \
  -F "profile=contract"

TIP

When in doubt, omit the profile parameter and let auto-classification choose. The keyword heuristic is fast and accurate for well-typed documents.