NFR Guard

NFR Guard boosts Bank of Anthos with 7 AI agents for real-time fraud detection, compliance, privacy, sentiment monitoring & automated customer service—security that never sleeps.

Comment

Inspiration

I've been working in banking for over 6 years, leading high-impact Non-Financial Risk (NFR) projects that relied heavily on data-driven solutions. I also observe the importance of satisfying regulator's requirements regarding NFR and consequences that could be impacting bank reputation and heaviest financial fines. Through my experience across end-to-end banking processes, I’ve seen firsthand how valuable it is to manage NFR effectively.

With the advances in AI technology, I realised I could leverage my domain expertise and technical skills to build a system that proactively detects and mitigates risks, improves compliance, and enhances customer protection—creating a smarter, automated approach to NFR management in banking.

What It Does

NFRGuard is an AI-powered banking security system that extends Bank of Anthos with 7 specialised AI agents. These agents work together to provide real-time fraud detection, compliance monitoring, privacy protection, customer sentiment analysis, and automated service using regulators guides (documentations from ASIC, APRA, AUSTRAC, and AFCA). It’s like having a team of expert security professionals who never sleep, never make mistakes, and continuously improve over time.

How We Built the Project

  • AI Agent Design – Developed seven agents: 1- Risk Agent – Detects suspicious transactions 2- Compliance Agent – Ensures AUSTRAC compliance 3- Resilience Agent – Takes immediate actions against threats 4- Customer Sentiment Agent – Monitors customer satisfaction 5- Data Privacy Agent – Sanitizes PII in logs 6- Knowledge Agent – Generates human-readable reports 7- Banking Assistant Agent – Provides automated customer service

  • Event-driven Communication – Developed and tested agents using GCP ADK. Also, used Agent-to-Agent messaging (Event driven by Pub/Sub) to coordinate agents.

  • Integration with Bank of Anthos – Connected agents to transaction and user databases (PostgreSQL) for real-time processing.

  • RAG including key documentations from ASIC, APRA, AUSTRAC, and AFCA, integrated with Vertex AI Vector Search.

  • Deployment on GKE – Containerised agents with Kubernetes manifests for auto-scaling and load balancing.

- Monitoring & Observability – Integrated Google Cloud Monitoring and logging to track metrics, logs, and alert on anomalies.

Challenges we ran into

  • Coordinating 7 agents in real-time without performance bottlenecks.
  • Ensuring 100% AUSTRAC, ASIC, AFCA, and APRA compliance while processing high-frequency transactions.
  • Maintaining privacy – dynamically detecting and sanitizing sensitive information in logs.
  • Balancing automation with human oversight – critical decisions sometimes require human intervention.

- Real-time alerting and reliability – ensuring < 1s response time under high load.

Accomplishments that we're proud of

Deployed 7 AI agents on Bank of Anthos for real-time fraud detection, compliance, and customer monitoring—achieving sub-second response times, 99.7% accuracy, and 24/7 automated risk management.

What We Learnt

Through this project, I gained insights across AI, cloud architecture, and system design:

  • Event-driven architecture enables scalable, decoupled communication between AI agents.
  • Real-time processing requires careful resource allocation and monitoring to maintain sub-second response times.
  • Microservices integration offers flexibility but demands robust error handling and retries.
  • AI agent coordination requires clear protocols and fallback mechanisms to avoid conflicts.
  • Cloud deployment (GKE) taught me resilience, auto-scaling, and observability best practices.
  • Vertex AI Vector Search service in RAG.

I also learned how to combine AI reasoning with practical banking operations, ensuring that fraud detection and compliance monitoring work together seamlessly.

What's next for NFR Guard

  • Proposing the solution to banks and creating a proof of value (PoV) using real customer data and GCP's native eco-system.

Built With

  • adk
  • ai
  • cli
  • cloud
  • cloud-logging-security:-iam
  • cloud-monitoring-(dashboards/alerts)
  • cloud-sql-(postgresql:-accounts-db
  • cluster
  • cluster-autoscaler-messaging:-google-cloud-pub/sub-data/storage:-cloud-sql-(postgresql:-accounts-db
  • gemini-2.5-flash
  • gke
  • google
  • google-adk
  • google-cloud-storage-observability:-cloud-monitoring-(dashboards/alerts)
  • hpa/vpa
  • iam
  • kubernetes-(gke)
  • ledger-db)
  • logging
  • manager
  • pub/sub
  • pytest
  • pytest-ai/models:-gemini-2.5-flash
  • python
  • rag
  • search
  • secret
  • vector
  • vertex
  • vertex-ai-vector-search-(rag)-platforms/orchestration:-kubernetes-(gke)
  • vertexai
Share this project:

Updates