Home About Technology Team Contact Us
AI-Native Financial Crime Prevention

Stop Financial Crime
Before It Settles

GPU-accelerated fraud detection and AML compliance infrastructure for banks, e-wallets, and fintech platforms across the Philippines and Southeast Asia. Sub-50ms decisions. Regulatory-grade accuracy.

<50ms
Inference latency per transaction
99.3%
Detection accuracy on PH data
10M+
Transactions analyzed daily
BSP
Aligned with PH AML mandates
GEMINISHIELD INFERENCE PIPELINE 01 EVENT INGESTION REST API / Stream — <1ms receive 02 FEATURE EXTRACTION Velocity · Graph · Behavioural 03 TRITON INFERENCE TensorRT · CUDA · 3 models parallel NVIDIA GPU 04 RISK DECISION Pass · Flag · Hold — <50ms total 05 FEEDBACK LOOP Labels → GPU retrain → Accuracy++ LIVE MONITORING TRANSACTIONS TODAY 2,847,391 ▲ 12.4% vs yesterday FRAUD BLOCKED 1,204 ₱8.3M exposure prevented AVG INFERENCE 38ms TensorRT optimized MODEL ACCURACY 99.3% 30-day rolling window Transaction Volume — 24hr Live Alert Queue TXN-884720 HIGH ₱240,000 38ms TXN-884718 MED ₱85,500 41ms TXN-884715 HIGH ₱610,000 35ms TXN-884712 CLEAR ₱12,200 29ms Entity Graph Model Health Precision Recall F1-Score STR Auto-drafted 11/14 LIVE

The Problem

Financial Crime Is Outpacing Legacy Detection

The Philippines processed over 1.5 billion digital payment transactions in 2023. This volume growth has created an equally fast-expanding attack surface that traditional rule-based systems cannot counter in real time.

BSP Circular 1140, AML Act amendments, and FATF pressure have placed unprecedented compliance obligations on every tier of financial institution.

  • 01

    Detection Latency

    Rule-based systems flag fraud after settlement. Chargebacks accumulate while static thresholds fail against adaptive attack patterns operating at digital payment velocity.

  • 02

    Compliance Overhead

    Manually reviewing Suspicious Transaction Reports and Customer Due Diligence files consumes analyst hours and produces inconsistent output under regulatory examination.

  • 03

    Data Fragmentation

    Transaction data, identity signals, device fingerprints, and behavioral patterns exist in silos. No unified model connects them for holistic real-time risk scoring.


The Platform

GeminiShield: One Platform for Financial Crime and Compliance

An end-to-end financial crime detection and compliance automation platform built on GPU-accelerated deep learning. Unified transaction monitoring, entity resolution, behavioral biometrics, and regulatory reporting in a single API-first architecture.

Real-Time Transaction Scoring

Every transaction scored against a multi-layer graph neural network in under 50 milliseconds. Models optimized with NVIDIA TensorRT and served via Triton Inference Server for consistent throughput at any volume.

Entity Resolution Engine

Proprietary knowledge graph links accounts, individuals, devices, and behavioral patterns to surface hidden beneficial ownership structures and mule networks invisible to rule-based systems.

Automated Compliance Reporting

STR generation, Customer Due Diligence documentation, and AMLC submission workflows automated using AI-assisted classification trained on Philippine regulatory templates.

Behavioral Biometrics Layer

Device telemetry, session patterns, and interaction velocity analyzed continuously to detect account takeover and synthetic identity fraud without adding friction to legitimate customers.


Who We Serve

Built for Every Layer of the Philippine Financial System

Rural and Thrift Banks

Affordable compliance infrastructure scaled to smaller transaction volumes, enabling BSP-compliant AML programs without dedicated in-house data science teams.

E-Wallet and Fintech Platforms

Real-time fraud scoring via REST API with sub-50ms SLA, built for the transaction velocity of major e-wallet networks operating across the Philippines.

Universal and Commercial Banks

Enterprise deployment with dedicated model instances, custom entity resolution graphs, and hybrid configurations for strict data sovereignty requirements.

Our Story

We Built What the Philippine Financial System Actually Needed

Gemini Tech Innovations was founded in 2023 in Quezon City, NCR, Philippines, with one clear objective: to give Philippine financial institutions access to GPU-accelerated compliance and fraud detection infrastructure without prohibitive implementation costs or lengthy deployment timelines.

Our founding team observed firsthand how rapid digitisation of Philippine payments created a systemic vulnerability. Institutions were onboarding millions of new digital users while fraud and compliance infrastructure remained fundamentally unchanged from the pre-digital era.

We built GeminiShield to close this gap. Today, Gemini Tech Innovations operates a team of 16 engineers, data scientists, and compliance specialists across model development, platform infrastructure, and client integration.

Our Data Advantage Is Not Replicable

GeminiShield's detection accuracy is built on a proprietary dataset of Philippine financial transaction patterns, entity behavior profiles, and labeled fraud incidents accumulated since our founding.

Developed through direct partnerships with financial institutions and curated by our compliance team, this dataset represents a data moat that external competitors cannot acquire through API access or public sources. Every analyst decision in the platform becomes a training signal, compounding our detection advantage over time.


Mission

To make AI-native financial crime prevention accessible to every institution in the Philippine financial system

Regardless of size, so that the growth of digital finance benefits everyone and exposes no one.

Vision

A Southeast Asian financial ecosystem where fraud is detected and neutralised in real time

Where compliance is automated by design, and institutions compete on customer experience rather than risk exposure.


What Drives Us

Our Values

Accuracy Over Speed

Every model we deploy is evaluated against Philippine-specific financial behavior data before production inference. We do not release models that have not cleared our internal accuracy thresholds.

Regulatory Fluency

Our compliance team maintains continuous alignment with BSP circulars, AMLC guidelines, and FATF recommendations. Clients should never have to translate regulation into product requirements themselves.

Infrastructure Discipline

We run on enterprise-grade cloud infrastructure because reliability, data residency, and uptime SLAs are non-negotiable in financial services. We do not cut infrastructure corners.

Local Knowledge as Moat

Our proprietary training datasets reflect Philippine transaction behavior, fraud patterns, and regulatory language. This cannot be replicated by a global vendor entering the market.

Under the Hood

GPU-Accelerated Deep Learning Infrastructure for Financial Crime at Scale

The GeminiShield platform is built on a multi-stage deep learning pipeline that processes transaction data, behavioral signals, and entity relationship graphs in real time. Every component is optimized using NVIDIA CUDA, TensorRT, and Triton Inference Server on dedicated GPU compute infrastructure.


NVIDIA Acceleration Stack

NVIDIA-Powered Inference at Transaction Speed

NVIDIA CUDA

Graph neural networks and transformer-based sequence models are trained using NVIDIA CUDA on dedicated GPU compute instances. CUDA-optimized frameworks reduce iteration cycles and enable continuous model improvement as new fraud patterns are identified in production data.

NVIDIA TensorRT

All inference models are compiled and optimized using NVIDIA TensorRT before deployment. TensorRT reduces per-transaction inference latency by up to 4x versus baseline models, enabling the sub-50ms scoring SLA that real-time payment fraud detection requires.

Triton Inference Server

Production model serving is handled by NVIDIA Triton Inference Server, which manages concurrent model instances, dynamic batching, and multi-model pipeline orchestration across fraud scoring, entity resolution, and behavioral biometrics models simultaneously.

Graph Neural Networks

Entity resolution and fraud network detection models are built on GNN architectures that process account relationship graphs, device fingerprint clusters, and behavioral similarity networks, trained on CUDA-enabled GPU compute using proprietary Philippine transaction datasets.


Inference Pipeline

The GeminiShield Scoring Pipeline

Every transaction passes through a five-stage pipeline delivering a composite risk decision in under 50 milliseconds.

01

Event Ingestion

Transaction events arrive via real-time streaming or synchronous REST API, carrying transaction metadata, device signals, session context, and entity identifiers.

02

Feature Extraction

A real-time feature engineering layer computes velocity features, graph-based entity similarity scores, and behavioral deviation metrics against historical baseline profiles.

03

Multi-Model Scoring via Triton

Triton Inference Server routes each transaction through three parallel model paths: an anomaly detection model, a fraud pattern classifier, and a GNN entity risk scorer. Results aggregate into a composite risk score.

04

Decision and Action

The composite score triggers automated hold, flag, or pass decisions in under 50ms. High-risk transactions are routed to the case management queue with pre-populated STR draft content for analyst review.

05

Continuous Feedback Loop

Analyst decisions on flagged transactions are captured as labeled training examples and queued for the next GPU retraining cycle, continuously improving model accuracy over time.


Full Stack

Technology Stack

NVIDIA CUDA NVIDIA TensorRT NVIDIA Triton Inference Server Graph Neural Networks Transformer Models GPU Compute Clusters Hyperscaler Cloud Infrastructure Real-Time Stream Processing Encrypted Cloud Storage Private Cloud Networking Python PyTorch Apache Kafka REST API PostgreSQL

The People

The Team Behind GeminiShield

16 engineers, data scientists, and compliance specialists with backgrounds in financial technology, cloud infrastructure, regulatory compliance, and applied machine learning. Headquartered in Quezon City, NCR, Philippines.

JP
Jerlyn Patron
Founder & Chief Executive Officer

Jerlyn Patron founded Gemini Tech Innovations in 2023 with the goal of bringing GPU-accelerated financial crime prevention to the Philippine fintech and banking sector. She leads company strategy, client partnerships, and product direction, combining expertise in financial technology with a deep understanding of the regulatory landscape governed by the Bangko Sentral ng Pilipinas. Under her leadership, Gemini Tech Innovations has built GeminiShield into a production-grade compliance and fraud detection platform serving clients across the Philippine financial system.

AN
Abanda Noel
Co-Founder & Lead Engineer

Abanda Noel leads the engineering and infrastructure function at Gemini Tech Innovations, overseeing the architecture of the GeminiShield platform, NVIDIA GPU compute integration, and cloud infrastructure operations. With a background in backend systems and applied machine learning, he drives the technical decisions that underpin GeminiShield's real-time inference performance and reliability at scale. He manages the team of engineers responsible for model deployment, API infrastructure, and real-time stream processing pipelines.


How We Are Organised

Our Team of 16

Model Development

Training, evaluation, and continuous improvement of fraud detection and compliance classification models on GPU compute infrastructure.

Platform Engineering

API infrastructure, real-time streaming pipelines, and cloud architecture ensuring uptime and data security across all client deployments.

Compliance and Regulatory

BSP alignment, STR workflow automation, AMLC reporting, and continuous monitoring of Philippine regulatory changes.

Client Integration

Technical onboarding, API integration support, and ongoing partnership management for banking and fintech clients across the Philippines.

Get in Touch

Let's Start a Conversation

Whether you are a financial institution evaluating GeminiShield, a technology partner exploring integration, or a researcher interested in AI-driven financial crime prevention, we welcome the conversation.

We respond to all inquiries within 2 business days.

Location
Blk 84 Lot 3 Corpus Christi St.,
Greater Lagro Subdivision,
Quezon City 1100, NCR, Philippines