Technical Foundation

✅ Pillar 1: Factflow (Knowledge Layer)

Purpose: Single source of truth that enables AI to know everything the bank knows

Capabilities:

• Transforms documents into queryable knowledge through streaming architecture
• Unifies all knowledge sources into a single, queryable intelligence system
• Eliminates duplicate work where every department builds their own data pipelines
• Enables real-time intelligence that keeps agents current with regulations, products, and markets

Scope:

• Complete DNB.no content (2000+ pages)
• Aino (customer-facing chatbot) knowledge
• Juno (customer care agent-facing chatbot) knowledge
• Elimination of contradictions and inconsistencies across sources

What Factflow Enables:

• Banking Knowledge: AI agents understand DNB's products, policies, and procedures
• Regulatory Compliance: Agents stay current with constantly changing regulations
• Market Intelligence: Real-time understanding of property markets, competitors, and trends
• Customer Context: Agents access complete, accurate information to serve customers
• Trust & Auditability: Every AI response can be traced to its knowledge source

✅ Pillar 2: Agent Prism (Orchestration Layer)

Purpose: Enable AI to actually do things, not just talk about them

Capabilities:

• Deploys and manages AI agents at scale with multi-model support
• Multi-step planning and execution (analyze → plan → execute → verify)
• Template-based agent creation (reduce "agent from scratch" to "agent from pattern")
• Automated validation and testing frameworks
• Compliance and security checks built-in
• Orchestration layer for multi-agent workflows

What Agent Prism Enables:

• Rapid Agent Deployment: Create specialized agents in hours instead of months
• KYC/AML Agents: Reduce customer onboarding from weeks/days to hours
• Mortgage Assistant: Guide customers through applications with integrated market data
• Service Agents: Handle routine inquiries without human intervention
• Risk Assessment: Real-time affordability and compliance checking
• Multi-Model Support: Break free from vendor lock-in, use any AI model (GPT, Claude, proprietary)

Breaking Free from the Past:

• No longer trapped in 2018-era chatbot technology
• Move beyond rule-based systems to true intelligence
• Accelerate deployment of intelligent banking agents that solve real customer needs
• Transform from navigating traditional interfaces to conversational, context-aware banking

📅 Pillar 3: Evaluation Layer (Future Roadmap)

Purpose: Quality assurance and performance monitoring for AI at scale

Planned Capabilities:

• Ensure AI responses meet DNB's standards for accuracy and compliance
• Track and validate regulatory adherence across all AI interactions
• Performance monitoring and continuous improvement
• Build confidence through measurable quality metrics

Why This Matters:

• PBA must maintain bank-grade quality and compliance at massive scale
• Regulators and customers need assurance that AI guidance is trustworthy
• Continuous learning requires measuring what works and what doesn't
• Enables DNB to deploy AI confidently across all banking domains

What's Different

• Team size: 1-2 people (human + AI digital twins) vs. traditional 20+
• Cycle time: Spec → build → iterate in days, not months
• Governance: Architecture and security guidance encoded as context (claude.md files) rather than committee approvals
• Capabilities: One person with AI augmentation can handle product spec, development, UX design, testing, documentation, and iteration

Evidence: We're Already Doing This

Radical AI team (8 people) built both Factflow and Agent Prism in 3 weeks of parallel development:

• Small team (2-4 engineers) working with AI augmentation
• Traditional approach: 10-15 engineers, 18-27 months
• 26-39X faster delivery
• 5X smaller team
• Superior quality (479 tests, 100% pass)