Document Policy Compliance Checker with Human-in-the-Loop

A beginner-friendly tutorial demonstrating how to build AI workflows with human oversight using LangGraph.

Overview

This notebook teaches you how to build an automated document compliance checker that reviews files for policy violations while keeping humans in the decision-making loop. You'll learn core LangGraph concepts through a practical, real-world example inspired by policy compliance work in government and enterprise organizations.

What You'll Build

An AI agent that:

• Scans documents for policy compliance violations
• Pauses execution when it finds potential issues
• Asks for human confirmation before continuing
• Resumes processing based on your decisions

You'll build this in two stages:

1. Pattern matching version - Fast, free, deterministic (no API key needed)
2. LLM-enhanced version - Smarter analysis using Claude (requires Anthropic API key)

Why Human-in-the-Loop?

Even smart AI systems need human oversight, especially for:

• Compliance and policy enforcement
• Ambiguous cases requiring context or judgment
• Situations where mistakes have real consequences
• Building trust in automated systems

This tutorial shows you how to build workflows where AI handles the tedious scanning work while humans make the final decisions.

What You'll Learn

• LangGraph fundamentals: State management, nodes, edges, and checkpointers
• Human-in-the-loop patterns: Using interrupt() and Command(resume=...) to pause and resume workflows
• Iterative development: Starting simple, then adding sophistication
• Trade-offs: When to use pattern matching vs. LLM analysis
• Real-world application: Why human oversight matters even with advanced AI

Prerequisites

• Basic Python knowledge
• Familiarity with Jupyter notebooks
• A conda environment (recommended) or pip

Optional: Anthropic API key for Part 2 (LLM-enhanced version)

• Get one at https://console.anthropic.com

Installation using conda

conda create -n langgraph-tutorial python=3.13
conda activate langgraph-tutorial
conda install -c conda-forge langgraph langchain-anthropic -y

Quick Start

1. Clone or download this repository
2. Open policy_compliance_checker.ipynb in Jupyter
3. Run the cells in order
4. Part 1 requires no API key - start here to learn the basics!
5. When ready for Part 2, add your Anthropic API key

Example Use Cases

This pattern works for many scenarios beyond document compliance:

• Legal/Compliance: Reviewing contracts, policies, or regulatory documents
• Content Moderation: Flagging content that needs expert review
• Quality Assurance: Checking code, documentation, or data quality
• Medical/Healthcare: Identifying cases requiring specialist attention
• Financial: Reviewing transactions or documents for approval

Repository Contents

• policy_compliance_checker.ipynb - Main tutorial notebook
• README.md - This file
• sample_documents/ - Example files that are created automatically when you run the notebook

Learn More

• LangGraph Documentation
• Anthropic API Documentation
• Explore LangGraph on anaconda.org

Extending This Project

Ideas for building on this tutorial:

• Add support for .docx, .pdf, or other file formats
• Implement batch processing for directories
• Generate compliance reports (PDF/HTML)
• Use persistent storage (databases) for checkpoints
• Add multi-user support with role-based review
• Create custom policy templates
• Integrate with document management systems or Slack

Acknowledgments

• Built with LangGraph
• Tutorial developed for the Anaconda community