This project demonstrates an enterprise-style Agentic AI workflow for casino revenue audit operations.
The application uses:
- OpenAI LLM tool calling
- deterministic reconciliation tools
- SOP-grounded Retrieval-Augmented Generation (RAG)
- workflow orchestration
- human-in-the-loop governance controls
The AI agent interprets auditor requests, determines which tools to execute, retrieves audit SOP guidance, identifies reconciliation variances, invokes approval workflows, and supports governed adjustment posting.
This project simulates how enterprise AI systems orchestrate workflows while maintaining financial controls and human approval requirements.
Gowrishankar Badanal Sivalingappa
LinkedIn: https://www.linkedin.com/in/gowrishankarbs
GitHub: https://github.com/gowrishbs84
Revenue audit teams in casino operations must reconcile financial data across multiple systems such as:
- SDS (Slot Data System)
- CMP (Casino Management Platform)
Traditional reconciliation processes are:
- manual
- spreadsheet-driven
- operationally expensive
- error-prone
- difficult to govern at scale
Auditors must:
- identify variances
- review audit policies
- validate adjustments
- capture approvals
- enforce governance rules
- rerun reconciliation workflows
This project demonstrates how an Agentic AI workflow can modernize revenue audit operations while preserving enterprise governance and financial controls.
- LLM-driven tool orchestration
- Deterministic eCash reconciliation tool
- SOP retrieval using RAG
- AI-generated audit recommendations
- Human-in-the-loop approval workflow
- Adjustment validation controls
- Secondary approval workflow for adjustments greater than $100
- Governance-aware adjustment posting
- Downloadable adjustment audit logs
- Session-aware workflow state management
- Streamlit UI for interactive workflow execution
This project demonstrates an enterprise Agentic AI architecture pattern.
Auditor Request ↓ LLM Agent ↓ Tool Selection ├── reconcile_ecash ├── retrieve_sop └── show_adjustment_module ↓ Python Orchestration Runtime ↓ Tool Execution ↓ Tool Outputs Returned to LLM ↓ LLM Audit Recommendation ↓ Human Approval Workflow ↓ Governance Validation ↓ Adjustment Posting
The AI agent reconciles SDS and CMP eCash data using deterministic reconciliation tools.
The LLM agent dynamically selects tools based on auditor intent.
Example tools:
- reconcile_ecash
- retrieve_sop
- show_adjustment_module
The AI assistant retrieves SOP guidance and generates governance-aware audit recommendations.
The adjustment workflow supports:
- human approval
- adjustment reason capture
- CMP adjustment validation
- governance enforcement
Adjustments greater than $100 require secondary approver validation before posting.
Unlike traditional AI-assisted workflows where Python controls the execution sequence, this project demonstrates:
LLM-driven workflow orchestration
The LLM determines:
- which tools should be executed
- when SOP retrieval is required
- when adjustment workflow should be invoked
- when human approval is required
- when governance escalation is necessary
The orchestration runtime executes the selected tools and returns results back to the LLM for reasoning.
Financial calculations are intentionally separated from the LLM.
The LLM does NOT:
- calculate revenue totals
- calculate variances
- modify financial values
All reconciliation calculations are performed by deterministic Python/Pandas tools.
This approach improves:
- auditability
- governance
- financial accuracy
- enterprise compliance
- operational trust
The project uses SOP-grounded Retrieval-Augmented Generation (RAG).
knowledge_base/ecash_audit_sop.txt
The LLM retrieves:
- audit SOP guidance
- approval rules
- governance requirements
- escalation policies
before generating recommendations.
The system enforces enterprise governance controls.
- Human approval required before posting adjustments
- Adjustment reasons must be captured
- Adjusted CMP values must match SDS values
- Secondary approver required for adjustments greater than $100
- Audit timestamps and committed user captured
- AI recommendations are advisory only
The LLM dynamically selects tools based on auditor intent.
| Tool | Purpose |
|---|---|
reconcile_ecash |
Compare SDS and CMP data and identify variances |
retrieve_sop |
Retrieve audit SOP guidance |
show_adjustment_module |
Trigger human approval adjustment workflow |
Please reconcile today's eCash audit.
Reconcile the eCash audit and retrieve the SOP guidance.
Show me the adjustments to be made.
Review variance records and open the approval workflow.
csv slot_location,gamingdt,ecash_in,ecash_out 101,2026-05-01,1200,500 102,2026-05-01,900,300 103,2026-05-01,1500,700 104,2026-05-01,2000,800
csv slot_location,gamingdt,ecash_in,ecash_out 101,2026-05-01,1200,500 102,2026-05-01,900,250 103,2026-05-01,1500,700 104,2026-05-01,1950,780
- Python
- Streamlit
- Pandas
- OpenAI API
- python-dotenv
py -m pip install -r requirements.txt
Create a .env file:
OPENAI_API_KEY=your_api_key_here
py -m streamlit run app.py
agentic-revenue-audit-assistant/ │ ├── app.py ├── requirements.txt ├── README.md ├── .env ├── .gitignore │ ├── data/ │ ├── sds_ecash.csv │ └── cmp_ecash.csv │ ├── knowledge_base/ │ └── ecash_audit_sop.txt │ ├── assets/ │ ├── reconciliation-screen.png │ ├── tool-calling-screen.png │ └── adjustment-module-screen.png │ └── architecture/ └── agentic-workflow-diagram.png
- Agentic AI workflow orchestration
- LLM tool calling
- Deterministic financial processing
- Retrieval-Augmented Generation (RAG)
- Human-in-the-loop governance
- Approval workflow orchestration
- Enterprise audit controls
- AI governance and compliance
- Operational AI design
- Workflow state management
- AI Solution Architecture
- Agentic Workflow Design
- Revenue Audit Automation
- AI Governance
- Tool-Oriented AI Design
- Enterprise Workflow Orchestration
- OpenAI Integration
- Python Development
- Streamlit Application Development
- Technical Program Management
- Financial Systems Thinking
- LangGraph orchestration
- Multi-agent coordination
- Vector database integration
- Conversational audit memory
- Real-time reconciliation engine
- Database integration
- Role-based access control
- Escalation management workflow
- Enterprise reporting dashboard
- Cloud deployment architecture
- Autonomous retry orchestration
- MCP integration
Gowrishankar Sivalingappa, PMP
AI TPM | Revenue Systems | AI Automation | Gaming Operations