Aegis
Every Transaction. Watched.
Real-time ML fraud scoring for Bangladesh banks and MFS operators. 3-layer cascade scoring, 80+ rules, BB Partner Network compliance, and Bangladesh-specific intelligence — all in under 50 milliseconds. Live at aegis.karitkarma.com.
A KaritKarma ProductScoring cascade
Every transaction passes through three layers in under 50ms.
Each layer is independent and stop-on-decision. The cascade short-circuits the moment a layer is confident enough to act, so deeper analysis runs only when it actually adds signal.
If layer 1 says allow with high confidence, layers 2 and 3 don't run. If layer 1 says deny with high confidence, the transaction is blocked before it touches the core banking system.
Rule cascade
- Hard rules first: sanctioned-party lists, velocity caps, country blocks.
- Deterministic and fully explainable. No ML in this layer.
- Stops on a high-confidence allow or deny before any model runs.
ML risk scoring
- Gradient-boosted decision trees, trained on 18 months of consortium data.
- Returns a risk score 0-1000 with per-feature attribution.
- Calibrated thresholds per bank, per channel, per BIN.
Human-in-the-loop
- Analyst review queue for ambiguous cases (risk score 600-800).
- Casework UI with full transaction history and customer context.
- Every decision feeds back into the rule and model registry.
Bangladesh-specific intelligence
Patterns the rest of the world doesn't see.
Off-the-shelf fraud platforms model US card-present and US e-commerce. Aegis ships with detectors purpose-built for the typologies that actually move money illicitly in Bangladesh.
Hundi corridor detection
Six high-risk divisions monitored. Pattern: split transactions to bypass reporting thresholds, routed through informal money transfer corridors.
MFS agent split fraud
Behavioral profiling on bKash and Nagad agent IDs. Pattern: agents structuring transactions just below KYC tiers across multiple customer accounts.
SIM swap takeover
Mobile-number change events correlated with first-login geography. Pattern: number ported, then large withdrawal initiated within 24 hours.
Synthetic identity
Document-photo correlation across new accounts. Pattern: same selfie or NID image attached to multiple identities.
Off-hours holiday spikes
Bangladesh holiday calendar built in. Pattern: transactions outside normal business hours during Eid, Pohela Boishakh, or government holidays.
Velocity anomaly per BIN
BIN-level velocity caps. Pattern: card BIN suddenly transacting at 10x its normal volume from a single merchant.
80+ rules curated by Bangladesh financial-crime analysts.
Every rule is versioned, auditable, and A/B-testable. Each one carries a tunable threshold so policy teams can dial sensitivity up or down without retraining a single model.
Velocity rules
3 of manySingle-account transaction count > 20 / hour
Card-not-present aggregate > BDT 2L / day
Cross-border outbound > BDT 5L / 24h
Identity rules
3 of manyFirst-time merchant + first-time card combination
Device fingerprint mismatch against 90-day history
IP geography distance > 500km from last successful login
Pattern rules
3 of manySequential round amounts (1k, 2k, 3k, ...)
Beneficiary account opened < 7 days before first inbound
Three failed-then-success pattern within 5 minutes
Each rule version is signed and timestamped — full audit trail for every flagged transaction. Tunable thresholds without retraining.
Regulatory mapping
Mapped clause-by-clause to Bangladesh Bank and BFIU.
Aegis isn't "compliance-adjacent." Each capability is mapped to a specific clause your auditors already cite — so the regulatory evidence package writes itself.
SIEM and monitoring
Continuous transaction-stream monitoring with structured event capture, mapped directly to Section 5 monitoring controls.
Incident response
Casework, escalation, and audit-log export aligned to the 72-hour incident-notification window.
STR and SAR triage
Threshold-aware triggers for Cash Transaction Reports, structuring, and Suspicious Activity Reports across all channels.
Digital transaction oversight
Sub-50ms scoring across MFS, card, and account-to-account flows so digital transactions are monitored inline, not in batch.
Protect Your Customers
9.3% MFS fraud rate. Zero existing solutions.
Bangladesh processes 12 billion MFS transactions annually with no dedicated fraud detection platform. Aegis fills that gap with production-grade AI that understands local fraud patterns.
Protect your customers from fraud.
Get a personalized walkthrough of Aegis with one of our specialists. No commitment required.