DeBERTa-v3-large · NLI Pipeline

Find what your
contracts disagree on.

Upload two legal documents. Our NLI model cross-compares every clause pair, surfaces contradictions by severity, and returns token-level evidence — so you know exactly where the conflict lives.

87.0%
Manual review reduction
94.2%
NLI precision
< 3s
Per comparison

Five-phase analysis,
end-to-end.

From raw file to token-level conflict evidence — each stage runs deterministically with no hallucination risk.

Multi-format ingestion

Extracts raw text from PDF and DOCX files server-side. No client-side parsing, no quality loss from OCR quirks.

Clause segmentation

Splits documents by sentence boundaries, producing clean clause units ready for pairwise comparison.

DeBERTa-v3 NLI

Microsoft's DeBERTa-v3-large fine-tuned on MNLI + SNLI. Three-way classification: entailment, neutral, or contradiction — per clause pair.

Severity scoring

Contradiction probability drives a four-band severity score (CRITICAL · HIGH · MEDIUM · LOW) so legal teams triage fast.

Token attribution

Integrated gradients surface the exact spans driving the prediction — no black box. You see why each conflict was flagged.

Clause type tagging

Each clause is automatically labelled: payment, liability, IP, termination, governing law, and 11 more — for structured downstream review.

Upload. Compare.
Resolve.

Drop two contracts below. Results include severity scoring, clause type labels, and token-level conflict attribution.

Demo mode — results are mock data. Set NEXT_PUBLIC_API_URL to connect the live backend.

Quick Test Workspace

Download our sample legal papers to review conflict patterns, or load them instantly to test.

Document A — drop here

PDF · DOCX · TXT

Document B — drop here

PDF · DOCX · TXT