Legal & Contracts

Profile: contract · Model: FLAGSHIP (gpt-5.6-sol) · Chunking: clause · Confidence: ≥ 0.75

Overview

Parse contracts, NDAs, service agreements, and procurement documents to extract parties, clauses, obligations, rights, monetary values, and key dates for contract review, obligation tracking, and compliance.

Extracted entities

Entity type	Examples
Party	Named parties with role (buyer, seller, licensor, licensee)
Clause	Specific contract sections with clause number
Obligation	What a party must do, with deadline
Right	What a party is permitted to do
MonetaryValue	Amounts with currency and payment terms
Date	Effective dates, expiry dates, payment dates

Extracted relationships

SIGNED_BY · OBLIGATES · GRANTS_RIGHT_TO · GOVERNS · REFERENCES

Clause-based chunking

The contract profile uses clause-based chunking instead of paragraph or section splitting. This respects legal section boundaries and keeps related provisions together in the same extraction window.

Why FLAGSHIP + high threshold

Legal documents are the highest-risk domain for LLM hallucination — particularly for party names. The contract profile uses:

• FLAGSHIP model (gpt-5.6-sol) for maximum extraction accuracy
• ≥ 0.75 confidence threshold to filter out uncertain extractions
• Confidence calibration to correct systematic LLM overconfidence

This combination minimises hallucinated party names — the most common LLM error on legal documents.

Example

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

{
  "entities": [
    {"type": "Party", "name": "Acme Corporation", "confidence": 0.83, "properties": {"role": "disclosing party"}},
    {"type": "Party", "name": "GlobalTech Ltd", "confidence": 0.79, "properties": {"role": "receiving party"}},
    {"type": "Obligation", "name": "Confidentiality obligation", "confidence": 0.76},
    {"type": "Date", "name": "2026-01-15", "confidence": 0.81, "properties": {"context": "effective date"}},
    {"type": "MonetaryValue", "name": "$500,000", "confidence": 0.77, "properties": {"context": "liquidated damages"}}
  ],
  "relationships": [
    {"source": "Acme Corporation", "target": "GlobalTech Ltd", "type": "SIGNED_BY"},
    {"source": "Confidentiality obligation", "target": "GlobalTech Ltd", "type": "OBLIGATES"}
  ]
}

Typical workflow

Contract upload (PDF, DOCX)
    ↓
DocIntel (contract profile) → parties, clauses, obligations, dates
    ↓
Purple8 Graph / JSON API → structured contract data
    ↓
Contract review tool → obligation tracking, expiry alerts, compliance