This project demonstrates how enterprise organizations can use AI agents to automate fraud detection and compliance investigation workflows using a modern Lakehouse + AI architecture.
The platform integrates Azure AI Foundry with a Databricks Lakehouse to analyze financial transactions, detect suspicious activity, generate investigation reports, and assist fraud analysts with AI-driven insights.
The system simulates a multi-agent fraud investigation workflow where different AI agents collaborate to evaluate financial transactions and produce compliance recommendations.
Data Generation Layer
- Synthetic financial transaction generation
- Customer profile simulation
Data Platform Layer
- Databricks Lakehouse
- Transaction datasets
- Data processing pipelines
AI Layer
- Azure AI Foundry
- GPT-4.1-mini reasoning model
AI Agent Layer
- Customer Data Agent
- Risk Analyzer Agent
- Compliance Report Agent
- Fraud Alert Agent
Application Layer
- Streamlit Fraud Investigation Dashboard
The Streamlit dashboard allows investigators to:
- analyze transactions using AI
- generate fraud risk assessments
- investigate suspicious activity
- produce compliance investigation summaries
Example investigation scenario:
Customer ID: C22 Transaction Amount: 5600 Location: Dubai Merchant: Luxury Electronics
The AI system analyzes behavioral anomalies and potential fraud indicators.
The AI generates:
- fraud risk score
- behavioral anomaly detection
- fraud explanation
- recommended investigation action
| Category | Technology |
|---|---|
| AI Platform | Azure AI Foundry |
| Data Platform | Databricks Lakehouse |
| AI Model | GPT-4.1-mini |
| Application | Streamlit |
| Programming | Python |
| Observability | OpenTelemetry |
| Data Storage | Lakehouse Tables |
Retrieves and analyzes customer transaction history.
Evaluates financial transactions for fraud indicators.
Generates investigation summaries and compliance reports.
Triggers alerts when suspicious transactions are detected.
Example query used in the system:
Analyze this transaction for possible fraud.
Customer ID: C22
Transaction Amount: 5600
Location: Dubai
Merchant: Luxury Electronics
Previous transactions: London
The AI returns:
- fraud risk score
- explanation of suspicious behavior
- recommended investigation action
pip install streamlit openai databricks-sql-connector python-dotenv
streamlit run ui/fraud_ai_ui.py
The application will start at:
http://localhost:8501
compliance-agent-platform
│
├ notebooks
│ ├ generate_fraud_transactions.ipynb
│ ├ customer_data_agent.ipynb
│ ├ risk_analyzer_agent.ipynb
│ └ compliance_report_agent.ipynb
│
├ ui
│ └ fraud_ai_ui.py
│
├ docs
│ └ images
│ ├ thumbnail.png
│ ├ architecture.png
│ ├ fraud_ai_dashboard.png
│ ├ fraud_transaction_analysis.png
│ └ ai_fraud_explanation.png
│
└ README.md
- real-time transaction streaming
- automated fraud scoring pipelines
- enterprise AI agent orchestration
- investigation workflow automation
- enterprise monitoring & logging
Harshit Tripathi Lead Data Engineer
Portfolio www.harshittripathi.com
GitHub https://github.com/harshitboots