01BDT 12.4 LC
Bangladesh banks processed BDT 12.4 lakh crore in digital transactions in 2025. Fraud losses rose 34% YoY. Most banks still run detection on overnight batches.
AEG-OS / BANGLADESH / v2026.04
The Decisioning Brainfor Bangladesh Banking.
Every transaction scored. Every typology caught. Every STR drafted. In your data center, under your control.
- <50mslatency
- 99.55%accuracy
- 80+rules
- PQCencryption
LIVE·mule ring detected·Dhaka·3 hops·BDT 4.85 Cr·17m
01 / THE PROBLEM
Fraud moves faster than controls.
The loss events of 2024–2026 across NPSB, BEFTN and mobile channels share a single pattern: credentials were legitimate, sessions were clean, devices were known. The fraud was in the movement of money.
0220%
80% of attack vectors are cross-channel — browser fingerprinting alone covers 20%.
034–7 days
BFIU reporting takes 4–7 days manually per STR. You have 30. Stack math is unforgiving.
02 / THE ARCHITECTURE INSIGHT
Device fingerprinting watches the door. AEGIS watches the money move.
Four layers of defense. Your competing vendors are pitching one. AEGIS is the transaction-layer brain — with the regulatory and consortium layers wired in.
01SOLVED
Channel & Identity Layer
Wenme · BioCatch · ThreatMetrix
Catches
Authentication, session integrity, device posture, bot detection. Confirms who is at the keyboard.
Blind to
Legitimate credentials moving stolen money. Insiders. Mule accounts on first-party devices.
02AEGIS
Transaction Decisioning Layer
AEGIS — purpose-built for BD
Catches
Behavioral DNA, graph-based mule detection, real-time scoring across every channel, counterfactual explainability. Decides whether the money should move.
Blind to
What bots look like on the login page. (Channel layer handles that.)
03BFIU
Regulatory & Reporting Layer
AEGIS + BFIU goAML · BB PN
Catches
Automated STR / CTR generation, BFIU goAML XML filing, sanctions, PEP lists, BB Partner Network telemetry.
Blind to
First-occurrence novel fraud — regulators tell you after the loss.
04NETWORK
Consortium Intelligence Layer
AEGIS Consortium · privacy-preserving
Catches
Cross-bank mule signal, shared typologies, blinded graph joins. Every participating bank gets stronger when any one is attacked.
Blind to
Bank-private insider patterns — by design, never left the DC.
You need every layer. We’re the one most banks are missing.
03 / Capabilities
Ten instruments for one hard job.
AEGIS ships with ten purpose-built capabilities. Each one is independently demo-ready. Together they are the decisioning brain.
AEG-01
Behavioral DNA
Per-customer 90-day embedding + real-time deviation score
AEG-02
Mule Ring Graph
Apache AGE Cypher + force-directed viz — circular, fan-out, fast-through
AEG-03
AI Investigator Copilot
Natural-language case investigation over graph, vector, and SQL
AEG-04
Auto-STR Generator
LLM-drafted BFIU STR + PDF in under 3 seconds
AEG-05
AML Typology Pack
Structuring, layering, smurfing, TBML, PEP cascade, shell, round-trip, rapid
AEG-06
Counterfactual Explainer
“Would approve if amount under 45k, OR if from known device.”
AEG-07
Consortium Intel
Cross-bank mule registry + blinded signal feed. Every attacked bank makes every bank safer.
AEG-08
Live Threat Heatmap
Real-time Bangladesh map — divisional pulses, cross-region flow lines
AEG-09
Wenme Session Signals
Device trust, geo, IP risk, MFA method — ingested as scoring features
AEG-10
Post-Quantum Crypto
CRYSTALS-Kyber / Dilithium · AES-256-GCM · SHA-3. Future-proof by default.
10 / 10 shippedEach module exposes a stable gRPC contract — one operator console, one audit trail.
03 / YOUR DATA CENTER
On-prem by design. Data sovereignty preserved.
No cloud egress. No PII off-site. Consortium participation is blinded by private-set intersection. Your data stays in your building.
THE STACK
A read-only Go CBS connector streams Temenos events over mTLS-gRPC. The Python intelligence layer scores every transaction across rules, gradient-boosted models, vector anomaly and graph walks. A .NET 10 platform gateway exposes CQRS-style APIs to your analyst console.
THE STORE
PostgreSQL 18 with pgvector, AGE and TimescaleDB is the one store — embeddings, graph, time-series, and ledger — with a 10-year immutable audit trail.
THE WIRE
Kafka in KRaft mode moves decisions between services at sub-ms p99. Post-quantum cryptography everywhere on the wire.
BUILT WITH
- Go 1.23
- Python 3.12
- .NET 10
- PG 18
- Kafka 4.2
- CRYSTALS-Kyber
- CRYSTALS-Dilithium
- AES-256-GCM
- SHA-3
REFERENCE · Prime Bank Data Centre · Dhaka
04 / REGULATORY
Bangladesh Bank Partner Network compliance deadline
31 December 2026
Automated STR filing, typology-based screening, consortium-grade mule detection — mandated across every retail and corporate channel. Manual compliance teams will not meet this deadline.
AEGIS is built to get you there.
Read the compliance mapTIME REMAINING
00Days
00Hours
00Mins
00Secs
remaining to roll out detection & reporting across the estate.
05 / TIMELINE
Discovery to production in fourteen weeks.
Banks do not deploy decisioning brains in a sprint. Five phases. Every phase has a written exit criterion and a joint sign-off. Week-14 go-live leaves 44 days of buffer before the 31 December deadline.
WEEK 02
Kickoff gate
Architecture & security review signed by CIO + CISO.
WEEK 05
Connector gate
Connector in bank DC streaming live Temenos events. No CBS writes.
WEEK 08
Shadow gate
Shadow decisions reconciled vs existing rule engine. P&R report.
WEEK 11
Pilot gate
Mobile App in active blocking mode. First STR auto-filed.
WEEK 14
Go-live
All channels in production. BB Partner Network telemetry enabled.
06 / THE DECISION
Three sentences. One decision.
01 / 03
Browser fingerprinting shows you a door-camera.
02 / 03
AEGIS shows you the whole building, every room, every hand that touches the money.
03 / 03
Your data stays in your DC. Our intelligence is at your service.
NEXT STEP
Pilot kickoff — 72 hours post-sign.
- Joint technical design workshop — AEGIS + your architecture team.
- Temenos connector security review scheduled with CISO office.
- BFIU liaison briefed on goAML submission path.
- Pilot channel selection — Mobile App recommended.