📄 Local Contract Triage (Multi-Pass AI)

A privacy-first, fully local web application designed to automatically scan, extract, and audit legal contracts for potential red flags. Built with Python and Streamlit, this tool leverages open-source Large Language Models (LLMs) to ensure your sensitive legal documents never leave your machine.

🌟 Overview

Reviewing contracts is often tedious and error-prone, but uploading sensitive agreements to cloud-based AI tools poses a massive security risk. Local Contract Triage solves this by running powerful AI models entirely on your local hardware.

By utilizing a Multi-Pass AI Architecture, the application minimizes hallucinations and ensures that any identified “Red Flags” are backed by exact, verbatim quotes from the original document.

✨ Key Features

  • Privacy First: 100% local processing. No data is sent to external APIs (OpenAI, Anthropic, etc.).
  • Multi-Pass AI Architecture: Splits the AI’s workload into smaller, specialized tasks (Extraction vs. Auditing) for significantly higher accuracy.
  • Verbatim Sourcing: The AI is constrained to provide exact quotes from the contract to prove its risk analysis.
  • Instant PDF Parsing: Rapidly extracts raw text from complex PDF documents using PyMuPDF.
  • Clean UI: An intuitive, responsive interface built with Streamlit.

🧠 The Multi-Pass Architecture

Instead of asking an LLM to read a massive contract and find issues in one go (which often leads to missed clauses or hallucinations), this application uses a two-step pipeline:

  1. Pass 1: Pure Extraction (The Gatherer)

    • Objective: Act as an objective data extractor.
    • Action: Scans the text strictly for clauses related to automatic renewals, early termination, and limits of liability.
    • Output: Returns only the verbatim text. No analysis.

  2. Pass 2: Risk Evaluation (The Auditor)

    • Objective: Act as a senior corporate paralegal.
    • Action: Reviews the narrowed-down clauses from Pass 1 and evaluates them for predatory risks or hidden costs.
    • Output: A formatted Red Flag report detailing the risk category, an analysis of why it’s harmful, and the verbatim quote as proof.

🛠️ Tech Stack

Component Technology
Language Python
Frontend Streamlit
PDF Parsing PyMuPDF (fitz)
Local AI Engine Ollama
LLM Model qwen2.5-coder:14b (Customizable based on your local hardware)

🚀 Getting Started

Prerequisites

  1. Python 3.8+ installed on your machine.
  2. Ollama installed and running in the background. Download Ollama here

Installation

  1. Clone the repository:

    git clone https://github.com/sarun1220/ai-contract-analyzer.git
cd ai-contract-analyzer

  2. Install the required Python packages:

    pip install streamlit pymupdf ollama

  3. Pull the Local LLM via Ollama: Note: This downloads the 14-billion parameter Qwen coder model. Ensure you have sufficient RAM/VRAM.

    ollama pull qwen2.5-coder:14b

Running the App

Start the Streamlit server:

streamlit run app.py

Navigate to the localhost URL provided in your terminal, upload a PDF contract, and click Scan Contract to see the magic happen!

Built as a Proof of Concept (PoC) to demonstrate secure, locally hosted AI workflows in the legal tech space.