Korastratum Agentic AI
Korastratum Agentic AI is a platform of 22 autonomous agents that observe, decide, and act across your banking operations. Every agent follows the same Observe-Decide-Act pattern, operates in shadow mode first, and graduates to active mode only after proving it matches or exceeds human decisions.
What you can do
- Automate compliance workflows — Auto-investigate compliance cases, disambiguate PEP/sanctions matches, route escalations to the right specialist, and monitor elderly accounts for exploitation.
- Prevent fraud in real time — Detect account takeover, structuring, velocity breaches, and fraud rings with autonomous freeze/OTP/flag responses.
- Accelerate lending decisions — Pre-screen loan applications using credit scoring, transaction analysis, and DTI calculations for instant approve/reject.
- Optimize treasury and settlements — Monitor liquidity positions, predict funding shortfalls, and auto-generate settlement batches at end of day.
- Improve identity verification — Auto-resolve IDV manual review queues, correct OCR name-matching errors, and tune liveness challenge parameters.
- Manage corridor performance — Analyze remittance corridor metrics, optimize FX quotes, and recommend fee adjustments.
Three-tier architecture
Agents are classified by the intelligence tier they use:
| Tier | Label | Count | How it works |
|---|---|---|---|
| Tier 1 | Rule-based (Core) | 12 | Deterministic rules and threshold logic. Fast, predictable, auditable. |
| Tier 2 | ML-powered | 10 | Bayesian scoring, anomaly detection, and pattern recognition models. |
| Tier 3 | LLM-powered | — | Reserved for future agents requiring natural language reasoning. |
All tiers share the same Observe-Decide-Act lifecycle, the same shadow mode safety net, and the same audit trail.
Zero regression guarantee
Every agent starts in shadow mode. In shadow mode, the agent observes real data and makes decisions, but never acts. Its decisions are logged alongside human decisions for comparison. After a minimum 2-week observation period, if the agent's agreement rate meets the activation threshold (typically 85%+), it can be promoted to active mode.
This means:
- No surprise actions — agents never act until they've proven accuracy
- Full audit trail — every decision is logged with confidence, risk level, and reasoning
- Human comparison — shadow decisions are compared against actual human outcomes
- Per-tenant control — each tenant controls which agents are active, shadow, or disabled
Products covered
The 22 agents span 6 product areas:
| Product | Agents | Examples |
|---|---|---|
| Compliance | 6 | Elder detection, compliance monitoring, escalation routing, case investigation, match verification, caregiver monitoring |
| Fraud | 2 | Real-time fraud prevention, fraud ring detection |
| IDV | 3 | IDV decision automation, name matching, adaptive liveness |
| Lending | 1 | Lending decision pre-screening |
| Treasury & Settlements | 3 | Treasury liquidity, settlement automation, reconciliation |
| Corridors & Payments | 4 | Corridor performance, quote optimization, bill pay recovery, rewards optimization |
| Inventory | 1 | Branch operations and inventory management |
See the full Agent Catalog for every agent ID, tier, and description.
How it works
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Trigger │────▶│ Agent │────▶│ Decision │
│ (event/ │ │ Observe → │ │ Logged + │
│ manual/ │ │ Decide → │ │ Acted on │
│ schedule) │ │ Act │ │ (if active)│
└──────────────┘ └──────▲───────┘ └──────────────┘
│
┌──────┴───────┐
│ Registry │
│ + Config │
│ (per-tenant)│
└──────────────┘
- A trigger fires — a scheduled job, an event from another service, or a manual API call
- The agent runs its Observe-Decide-Act cycle using tenant-specific configuration
- The decision is logged to the audit trail; if the agent is in active mode and the risk is within threshold, the action executes automatically
- If risk exceeds the threshold, a human approval request is created in the workflow system
Next steps
- Quickstart — Enable and configure your first agent in 5 minutes
- Agent Overview — Browse all 22 agents with descriptions and tiers
- Shadow Mode — Understand the safety model before activating agents
- API Reference — Explore every endpoint