Aegis
Every Transaction. Watched.
Real-time AI fraud detection for Bangladesh banks and MFS operators. 3-layer cascade scoring, 80+ rules, and Bangladesh-specific intelligence — all in under 50 milliseconds.
A KaritKarma ProductScoring Architecture
Three layers. One decision. Under 50ms.
The cascade architecture routes each transaction through progressively deeper analysis, with early exit for clear-cut cases. 90% of transactions are resolved in under 2ms.
Layer 1: Rules Gate
<2ms80+ rules across 14 categories — velocity, geographic, device, AML, account takeover, mule detection, SIM swap. Instant approve or block for clear-cut transactions.
Layer 2: Fast Model
2-5msGradient boosting model trained on 265 features. Three-band output: approve below 15, block above 85, pass through the ambiguous middle to Layer 3.
Layer 3: Deep Ensemble
15-25msParallel ML components — GBM, vector brain, graph engine, LSTM — combined by a stacking meta-learner. Isotonic calibration maps to true fraud probability.
Decision Bands
FinalDynamic thresholds: APPROVE 0-30, REVIEW 31-60, STEPUP 61-85, BLOCK 86-100. Adjusts for night hours, high-risk MCCs, large amounts, and new accounts.
Core Capabilities
Built for Bangladesh. Ready for the world.
Purpose-built fraud intelligence combining global best practices with deep Bangladesh-specific knowledge — hundi corridors, MFS agent networks, SIM swap patterns, and regulatory compliance.
3-Layer Cascade Scoring
Rules gate in under 2ms, fast GBM model in 2-5ms, deep ensemble in 15-25ms. Early exit means 90% of transactions never reach the expensive layers.
80+ Detection Rules
14 rule categories covering velocity, geographic, device, AML, account takeover, mule detection, dormant accounts, and more. Per-bank threshold overrides.
Bangladesh Intelligence
Hundi corridor detection across 6 divisions, MFS agent behavioral profiling, SIM swap risk scoring, and Bangladesh holiday-aware transaction analysis.
Graceful Degradation
Four-level degradation (Normal, Elevated, High, Critical) based on p95 latency. System always returns a decision, even under extreme load.
Alert & Regulatory Engine
Band-based alerts via SMS, WhatsApp, push, and dashboard. Automatic CTR, structuring, and SAR triggers per Bangladesh Bank regulations.
Consortium Intelligence
Federated learning across participating banks. Cross-institution fraud pattern sharing without exposing raw transaction data. The SaaS moat.
Bangladesh Intelligence
Fraud patterns only locals understand.
Generic fraud systems miss Bangladesh-specific attack vectors. Aegis was built from the ground up with deep knowledge of local fraud typologies, regulatory requirements, and financial infrastructure.
Hundi Corridor Detection
Monitors 6 high-risk divisions (Chattogram, Brahmanbaria, Cumilla, Kushtia, Khulna, Bagerhat) for off-hours corridor activity, round-amount repetition, and cross-border remittance correlation.
SIM Swap Risk Scoring
5-level risk assessment (NONE through CRITICAL) based on swap recency, device age, transaction value, and synthetic identity detection via NID-linked account counts.
MFS Agent Profiling
Behavioral baselines for mobile money agents. Detects split transactions, float drain, collusion signals, and off-hours anomalies against established agent patterns.
Regulatory Compliance
Automatic CTR filing for cash transactions above BDT 10 lakh. Structuring detection at 80-99% of threshold. SAR triggers on accumulated high-severity rule breaches.
Deployment
Live in weeks, not months.
Aegis deploys as a lightweight agent in the bank data center with cloud scoring, or as a fully on-prem appliance for banks that require it.
Deploy connector agent
A lightweight Go binary installs in the bank DC, connects to the CBS, and streams transactions via encrypted gRPC. Zero changes to existing CBS infrastructure.
Shadow mode validation
Aegis scores every transaction in parallel with existing systems. Compare results, tune thresholds, and validate detection rates before going live.
Activate real-time scoring
Switch from shadow to active mode. Aegis scores transactions inline, publishes alerts via SMS, WhatsApp, and dashboard, and triggers regulatory reports automatically.
Continuous improvement
Analyst feedback loops retrain models. Consortium intelligence improves detection across all participating banks. Monthly performance reports track ROI.
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.