AMACS is an intelligent, multi-agent ecosystem built on CrewAI that automates the end-to-end financial auditing process. By orchestrating a team of specialized AI agents, AMACS transforms raw financial data into comprehensive, risk-assessed audit reports with human-level reasoning.
AMACS utilizes a sequential and hierarchical task execution model where agents hand off verified data to one another.
graph TD
User((User)) -->|Uploads Data| UI[Streamlit Interface]
UI -->|Triggers Crew| Orchestrator{CrewAI Orchestrator}
subgraph Agents
Orchestrator --> A1[Data Ingestor]
A1 -->|Clean Data| A2[Compliance Agent]
A2 -->|Policy Gaps| A3[Fraud Analyst]
A3 -->|Risk Flags| A4[Lead Auditor]
end
A4 -->|Synthesis| Report[Final Audit Report]
Report -->|Display| UI
AMACS is a multi-agent AI system built using CrewAI that automates financial auditing workflows including:
- Data ingestion (ETL)
- Compliance validation
- Fraud detection
- Risk-based audit reporting
User → Streamlit UI → CrewAI Agents → Data Ingestor → Compliance Agent → Fraud Analyst → Lead Auditor → Final Audit Report
- Python
- CrewAI (multi-agent orchestration)
- OpenRouter (LLM backend)
- Streamlit (UI)
- Pandas + DuckDB (ETL)
- ChromaDB (memory - upcoming)
- Multi-agent collaboration
- Explainable fraud detection
- Compliance rule engine
- Automated audit report generation
- Risk scoring system
git clone https://github.com/19Vermouth/amacs-audit-system.git
cd amacs-audit-systempython -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
streamlit run app.py
Create .env file:
OPENROUTER_API_KEY=your_key_here
- Compliance violations
- Fraud alerts with risk score
- Final audit report
- RAG for compliance documents
- Agent memory system
- Advanced fraud ML models
- Dashboard visualizations
Ishaan Kar