---
title: Contract Red Team
emoji: 🔍
colorFrom: yellow
colorTo: blue
sdk: gradio
app_file: app.py
pinned: false
license: mit
---

# Contract Red Team

## Question

How can an AI assistant review a contract while keeping evidence visible?

## System Boundary

This Space is a contract-risk triage tool. It is not legal advice. It is designed to surface clauses worth human review.

## Method

The app extracts text from a PDF, chunks the contract into clauses, searches for risk patterns, and uses model-assisted explanation to describe potential issues such as termination, non-compete, IP assignment, liability, and arbitration.

## Technique

This is evidence-first document triage. The system combines deterministic risk patterns with model explanation.

The design goal is to keep the clause visible. The model should not be a magic summarizer; it should point to text that a human can inspect.

## Output

The app returns a risk report with clause evidence, category labels, and suggested review questions.

## Why It Matters

High-stakes document AI must be evidence-first. The user should see why a clause was flagged, not merely receive a summary.

## What To Notice

The strongest outputs are tied to exact clauses. If the evidence is vague, the finding should be treated as weak.

## Effect In Practice

This workflow can help users prioritize review time by surfacing clauses that deserve attention before a detailed legal review.

## Hugging Face Extension

The Space can be expanded with a clause-risk dataset, model explanations, and evaluation on category precision and evidence quality.

## Limitations

Contracts are jurisdiction-specific and context-dependent. This tool should be used for education and triage only, followed by professional legal review.

## Run Locally

```bash
pip install -r requirements.txt
python app.py
```