NexusGuard Documentation & Investor Dossier
Operating under the philosophy "ML Assesses. Graphs Connect. AI Investigates. Humans Decide." Unifying sub-25ms dual machine learning inference, multi-hop financial graph forensics, millisecond behavioral biometrics, and autonomous AI agents across 212,749 evaluated transactions.
$3.2 Trillion in Global Financial Crime Meets 95%+ False Positives
Digital banking and Mobile Financial Services (bKash, Nagad, UPI, FedNow) operate in real-time, yet compliance and fraud detection remain trapped in legacy architecture:
Traditional rule-based engines (Actimize, Mantas) generate massive operational alert fatigue, costing banks billions in wasted human review.
Fraudsters drain stolen funds within 4 minutes, while legacy core banking batch jobs run overnight after the money has already left the rails.
Organized syndicates split funds across 12+ intermediary mule wallets and circular loops, completely invisible to single-account queries.
NexusGuard: Autonomous Financial Crime Intelligence OS
A synchronized, 4-tier defense engine that assesses risk in sub-25ms, maps topological mule rings in real time, and synthesizes audit-defensible SAR dossiers with explainable TreeSHAP reason codes.
Supervised LightGBM + Unsupervised Isolation Forest scored on rolling Redis sliding windows. Delivers instant risk probabilities with zero cold-start delay.
NetworkX topological centrality and cycle detection catches money mule syndicates, rapid structuring smurfs, and layering funnels across 4+ counterparty hops.
Millisecond keystroke cadence entropy and mouse flight tracking detects credential-stuffing and device-pairing account takeovers before transfers clear.
Specialized agent team compiles statutory Suspicious Activity Reports citing BFIU & FATF regulations, sealed into a cryptographic SHA-256 ledger.
Instant Rails + Regulatory Mandates + Frontier AI Convergence
1. Instant Rail Adoption
Mobile wallets handle hundreds of thousands of transactions per minute in emerging markets. Latency tolerance has dropped from hours to <50 milliseconds.
2. Central Bank Accountability
FATF Recommendations 10 & 16 and Bangladesh Bank guidelines require affirmative proof of non-discrimination, explainable AI reason codes, and auditable evidence trails.
3. Model Context Protocol (MCP)
Decoupling tabular ML and graph algorithms into standard MCP tools allows frontier LLMs to reason over financial topologies without exposing raw SQL or leaking customer PII.
Live Operational Scenarios Ready for Evaluation
Explore live forensic cases injected with synthetic ground truth directly inside the mission control center:
Hero Scenario: TXN-982173
3 failed PIN attempts, attacker hardware pairing, and transfer to mule WAL-5512.
ScamShield Mule Ring
Interactive Cytoscape visualization of 12 wallets aggregating into 2 cashout hubs.
Live Transaction Stream
Real-time multi-engine scoring with sub-25ms latency badges and risk filters.
Model Governance & SHAP
TreeSHAP global gain distributions and continuous population stability monitoring.
$28.5B Global AML Market with Explosive Asian MFS Demand
Global anti-money laundering software, transaction surveillance, and AI financial crime defense.
High-growth digital banking, Mobile Financial Services, and real-time payment rails across South & Southeast Asia.
Initial beachhead across 35 commercial banks, MFS operators (bKash, Nagad), and cross-border remittance providers.
Predictable Enterprise SaaS + Usage Tiers (82% Gross Margin)
Up to 500k monthly transactions
- ✓ Sub-25ms Dual ML Scoring
- ✓ Basic Rule Engine & Alerts
- ✓ Standard REST & Webhook API
Up to 5M monthly transactions
- ✓ NetworkX Multi-Hop Graph Forensics
- ✓ WalletGuard Behavioral Biometrics
- ✓ Autonomous LangGraph SAR Agent
- ✓ TreeSHAP Explainability Ledger
Unlimited transactions · On-Premise
- ✓ Local Ollama / vLLM Air-Gapped LLMs
- ✓ Zero PII egress guaranteed
- ✓ Custom SLM QLoRA fine-tuning
- ✓ 24/7 Dedicated Forensic Architect
Validated Across 212,749 Evaluated Transactions
Why NexusGuard Wins Against Legacy & Naive LLMs
| CAPABILITY | NEXUSGUARD | NICE ACTIMIZE | NETGUARDIANS | STANDALONE LLM |
|---|---|---|---|---|
| Inference Latency | <25ms (11.2ms avg) | >500ms / Batch | ~150ms | 1,500ms – 4,000ms |
| Graph Link Forensics | 4-Hop Real-Time BFS | Offline Batch Only | Partial 1-Hop | None (Context Limit) |
| Behavioral Biometrics | Keystroke & Jitter Entropy | No | No | No |
| Autonomous SAR Reports | LangGraph Agent (Cites FATF) | Manual Templates | Manual Templates | Unguided (Hallucinates) |
| Explainability Standard | TreeSHAP Reason Codes | Static Rules | Opaque Anomaly | Black Box Probabilities |
| Sovereign Air-Gapped Mode | Yes (Ollama / vLLM) | On-prem (Heavy) | Cloud only | Cloud API (PII Leak) |
Direct Tier-1 MFS Sales + Core Banking Integrator Channels
Direct engagement with the 3 largest mobile wallet operators in Bangladesh and Pakistan to replace brittle legacy threshold rules.
Pre-built adapters for Finacle, Temenos, and Mambu to enable zero-code drag-and-drop surveillance across conventional commercial banks.
Collaboration with central bank financial intelligence units (BFIU) for standardized SAR electronic filing APIs.
NexusGuard Core Architecture & Forensics Group
The interdisciplinary team of fraud forensic architects, distributed graph engineers, and AI safety researchers designing high-throughput trust infrastructure for mobile financial services.
Reyazul Islam
Chief Risk Officer
Clearance L4 · Super AdminPioneered sub-25ms dual machine learning scoring pipelines and local TreeSHAP reason-code attribution. Ex-central bank financial crime advisor.
Sadia Jannat
Senior AML Forensic Investigator
Clearance L3 · Forensic LeadSpecialized in FATF Recommendations 10/16 and Bangladesh Bank MFS regulations. Author of the autonomous LangGraph SAR dossier generator.
Mohammad Badol
Principal ML
Clearance L3 · ML OpsEngineered NetworkX multi-hop cyclic graph traversals, PageRank centrality, and real-time Isolation Forest zero-day anomaly detectors.
Rafi Mahmud
Risk Operations
Clearance L2 · SOC TriageDirects the 24/7 transaction stream intercept queue, human-in-the-loop approvals, and WalletGuard behavioral biometric hardware pairing validation.
The Global Trust Operating System for Autonomous Financial Rails
As artificial intelligence agents begin transacting autonomously across borders via FedNow, Pix, and regional MFS networks, manual compliance will collapse. NexusGuard is establishing the sovereign, explainable verification standard where every digital dollar is cryptographically authenticated, dynamically guarded, and legally defensible.
System Overview & Operational Roles
NexusGuard is architected as an asynchronous, decoupled, multi-engine platform engineered to handle burst loads up to 50,000 transactions per second with strict sub-25ms P99 inference latency guarantees.
Review live high-severity alerts, examine Cytoscape mule graphs, and sign off on automated wallet freezes.
Monitor Population Stability Index (PSI) drift, inspect TreeSHAP gain weights, and backtest shadow rule models.
Live System Capabilities & Verification State
End-to-End Technical Topology
Click any component node in the pipeline below to inspect its stack, performance SLA, and operational role.
LightGBM Gradient Booster + Isolation Forest + TreeSHAP
Simultaneously evaluates supervised fraud likelihood and unsupervised zero-day anomalies across 212k baseline distributions.
Sub-25ms Real-Time Ingestion to Ledger Seal
Follow the transaction lifecycle from raw edge arrival through dual inference, graph analysis, and immutable audit chaining.
Simultaneously generates supervised fraud probability and unsupervised anomaly scores. Computes TreeSHAP relative gains to formulate Reason Codes (RC-01 to RC-12).
Institutional-Grade Open-Source Stack
Next.js 16 + React 19
Turbopack, Server Components, Three.js WebGL spatial lattice, Cytoscape network links, and Tailwind CSS.
FastAPI + Async Python
Pydantic v2 schemas, SQLAlchemy ORM, Uvicorn ASGI, and OpenTelemetry trace spans.
LightGBM + NetworkX
Scikit-Learn Isolation Forest, TreeSHAP reason attributions, and BFS multi-hop cycle algorithms.
PostgreSQL 16 + pgvector
Redis 7 sliding windows, 1536-dim cosine embeddings for BFIU statutes, and SHA-256 audit ledger.
High-Throughput Scoring & Governance Endpoints
Real-time sub-25ms synchronous risk scoring. Evaluates dual ML models, checks sliding window velocity, and returns calibrated risk score with TreeSHAP reason codes.
curl -X POST "http://localhost:8000/api/v1/transactions/score" \
-H "Authorization: Bearer nxo_live_77f8a9..." \
-H "Content-Type: application/json" \
-d '{"transaction_id": "TXN-982173", "sender_wallet": "WAL-9821", "recipient_wallet": "WAL-5512", "amount": 18700.00, "channel": "MFS_APP"}'Fetches entity neighbor links, detected mule clusters, and circular wash loops for interactive graph visualization.
Sovereign Privacy & 212k Transaction Dataset Schema
Customer data is governed by zero-trust isolation. PII (National ID, Phone numbers, Device MAC hashes) is hashed with SHA-256 salting before ML vectorization or agent prompt injection.
Dual ML Scoring + TreeSHAP Reason Codes
Predictions are backed by local mathematical feature attribution. Every SAR report includes verifiable reason codes (e.g. RC-04: Velocity Surge >1400%, RC-08: New Hardware Hash Off-Hours).
Engineering Milestones (2026)
Sub-25ms Dual ML Core
212,749 transaction ingestion, TreeSHAP explainability, and WalletGuard behavioral biometrics.
Core Banking Connectors
Finacle and Temenos integrations, automated regulatory e-filing directly to BFIU portals.
Cross-Border Remittance Bridge
Cross-border graph link analysis connecting South Asia MFS rails with GCC remittance corridors.
Version 4.0.0 Production Release
Full population of 212k transaction dataset, calibrated risk scores (4.5k Critical, 2.6k High), 1-second masterclass skeleton loaders, and YC pitch deck integration.
Added Population Stability Index (PSI) drift tracking and Reason Codes (RC-01 to RC-12) local attribution.
LightGBM (v2.4.1), Isolation Forest (v1.9.0), and Ollama local daemons are actively verifying live transactions.