Compliance Hawk - AI Legal Auditor

Vercel
application badge
Created by team Compliance Hawk on November 23, 2025
IBMLlama 3.2

The Problem Enterprises hire thousands of employees annually, generating a massive volume of employment contracts. Reviewing these manually is slow (averaging 30 minutes per contract), prone to human error/fatigue, and costly. A single overlooked non-compliant clause—such as a "7-day notice period" where the law mandates "30 days"—can expose the company to significant legal risks and penalties. The Solution: Compliance Hawk Compliance Hawk is not just a document reader; it is an autonomous AI Agent designed to act as a Level-1 Legal Auditor. Unlike standard LLM wrappers, it combines the reasoning power of Generative AI with deterministic rule enforcement to provide 100% reliable audits. How It Works (Technical Architecture) We leveraged the IBM watsonx Orchestrate platform to build a seamless agentic workflow: Orchestration: IBM watsonx Orchestrate manages the conversation flow and skill routing. Intelligence: We utilized the Llama 3.2 model for its superior ability to extract specific clauses and reason against legal context. Custom Skills: A Python FastAPI backend (exposed via ngrok) handles the "Audit Contract" skill. Logic: The agent extracts key parameters (Salary, Notice Period, Role) and validates them against hard-coded Indian Labor Law standards. Key Features Risk Scoring: Assigns a binary Risk Score (0 for Safe, 100 for Critical) to eliminate ambiguity. Edge Case Handling: Smart enough to detect missing clauses (e.g., if a notice period isn't mentioned at all) and flag them as non-compliant. Instant Reasoning: Provides a clear legal explanation for every flag. Impact Compliance Hawk reduces the contract review lifecycle from 30 minutes to just 6 seconds per document, ensuring uncompromising compliance with zero manual oversight.

Category tags:
GithubPresentationDemo
Explore more applications

Toddle AI

Toddle AI is a privacy-first Android app that analyzes toddler walking videos on-device, checks capture quality, measures gait patterns from pose landmarks, and lets parents ask a local AI assistant to explain the result

Toddle AI

Qualcomm Snapdragon

Lantern: Model anything

Lantern turns a smartphone into an AI-powered 3D scanner. Using on-device AI with ExecuTorch, it generates accurate 3D meshes in real time without cloud processing, LiDAR, or external hardware, making 3D scanning fast, private, and accessible.

Lantern

Meta ExecutorchClaude CodeQualcomm Snapdragon

Memory Recall Assistant

A private, on-device memory assistant for cognitive decline. Using ExecuTorch on Snapdragon 8 Elite NPU, it identifies faces, transcribes conversations via Whisper, and generates warm recall cues via Llama 3.2 1B - fully offline, zero cloud.

Synapse Riders

Meta ExecutorchQualcomm SnapdragonWhisperLlama 3.2AI StudioAnthropic ClaudeGithub Copilot

SafeScreen AI

An on-device visual safety layer that helps protect people from explicit, abusive, and manipulated media before they interact with it.

SafeScreen AI

Qualcomm SnapdragonClaude Code

PrivateDoc

PrivateDoc is an offline, on-device visual assistant that uses ML Kit OCR, ExecuTorch, MiniLM, and Llama 3.2 to scan, index, and answer questions about private documents. Its strict RAG pipeline provides source-grounded answers. Powered by SnapDragon.

Edge Faithful

Qualcomm SnapdragonGithub CopilotMeta ExecutorchLlama 3.2

"For future development, focus on creating robust human-in-the-loop audit trails to ensure every suggested rewrite is fully explainable, traceable, and legally defensible before being applied in a real-world scenario. Using a domain-specific model like IBM Granite along with a corporate policy Knowledge Base (RAG) is the most effective and responsible way to approach high-stakes legal auditing, ensuring reliability beyond general-purpose tools. Loved the secondary agent to rewrite legal text."

avatar

Pavan Kumar Gondhi

Sr Vice President