Agreemind

AI-powered contract risk analysis across multiple legal domains.

We build specialized NLP models and an end-to-end analysis pipeline that automatically detects risky clauses in legal contracts — Terms of Service, Banking Agreements, NDAs, and general contracts.

🏗️ Architecture

Agreemind uses a multi-engine routing architecture: an intelligent router auto-classifies each document, then dispatches it to the best specialized engine:

📊 Models

Terms of Service (LexGLUE UNFAIR-ToS)

Fine-tuned on the LexGLUE UNFAIR-ToS benchmark. Evaluated on the official test set (1,607 samples).

LexGLUE Leaderboard: Legal-BERT (paper) = 96.0 μ-F1 / 83.0 m-F1. Our top models match or exceed this.

English Banking

Fine-tuned RoBERTa on 90 labeled US/UK consumer banking contracts (~4,337 clauses). Detects 9 risk categories.

Risk categories: Hidden Fees, Unilateral Rate/Terms Changes, Overdraft Penalties, Auto-enrollment, Data Sharing, Dispute Limitations, Account Freeze/Closure, Rewards Restrictions.

Turkish Banking

Fine-tuned BERT on manually labeled Turkish bank contracts.

NDA (Contract NLI)

NLI-based models trained on ContractNLI for 17 standard NDA provisions (3-class: Entailment / Contradiction / Not Mentioned).

🚀 Quick Start

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Terms of Service analysis
model_id = "Agreemind/lexglue-roberta-unfair-tos"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

text = "We may terminate your account at any time without notice."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=128)

with torch.no_grad():
    probs = torch.sigmoid(model(**inputs).logits).squeeze()

labels = ["Limitation of liability", "Unilateral termination", "Unilateral change",
          "Content removal", "Contract by using", "Choice of law", "Jurisdiction", "Arbitration"]

for label, prob in sorted(zip(labels, probs), key=lambda x: x[1], reverse=True):
    if prob > 0.5:
        print(f"  {label}: {prob:.3f}")

📎 Links

• LexGLUE Paper — ToS benchmark
• ContractNLI — NDA benchmark
• LexGLUE Dataset