Ricky-7-Yan/intelligent-audit-system

What it solves

AuditPilot is an enterprise-grade AI Agent workbench designed specifically for corporate audit delivery. It addresses the need for traceable, explainable, and auditable workflows in auditing—moving beyond simple chat interfaces to handle evidence collection, control mapping, risk assessment, and remediation loops.

How it works

The system uses a hybrid intent router to direct requests through a specialized agent chain (including Planner, Evidence, Control, Risk, Compliance, and Remediation agents). It integrates an agentic RAG system for grounded evidence retrieval and a memory system (Working, Episodic, and Profile) to maintain context across audit cycles. The workflow includes safety gates, reflection mechanisms, and human review to ensure high-risk or low-confidence outputs are verified by a person.

Who it’s for

It is built for corporate auditors and audit teams who need to automate repetitive tasks like evidence gathering and checking while maintaining strict compliance and traceability standards.

Highlights

• Agentic RAG: Features knowledge writing, chunking, and source citations with evidence quality gates.
• Evaluation Harness: Includes a release gate and baseline comparison to track agent and RAG performance.
• MCP-style Tools: Implements a tool system with schema definitions, permission declarations, and circuit breakers.
• Comprehensive Workspace: Manages the entire audit lifecycle from project initiation and control matrices to sampling plans and delivery packages.