Selected work
Enterprise & public sector · Banking

A shared biometric platform inside a private bank

The AWS foundation and server-side solution for biometric liveness and identity-photo matching at First Abu Dhabi Bank. Built for Payit onboarding on a fully private EKS cluster, with no internet ingress or egress.

  • Amazon EKS
  • Amazon MSK
  • VPC endpoints
  • Transit Gateway
  • CloudFormation
  • Lambda
  • KMS
  • MongoDB Atlas
  • React
  • Node.js
Read the case study

The engagement

I architected and implemented the AWS foundation and server-side solution for biometric liveness and identity-photo matching at FAB. The initial engagement focused on Payit wallet onboarding, with an architecture intended to support additional bank channels.

Conceptual view · not the production topology
Bank channels
Payit walletfirst channel
IBMB and Dubai Firstadded as the service expanded
Bank backend
API gatewayregistration journey stays with the bank
Private verification service
Liveness & face matchKubernetes on EKS
Protected dataencrypted fields
AWS servicesprivate endpoints

No direct route to the internet, in or out.

The delivery challenge

A shared service needs more than a working biometric integration. The infrastructure, application interfaces, network boundaries, and operational requirements have to work together across the channels that depend on it.

The platform had to fit the bank’s security and networking requirements and run on a fully private EKS cluster, with no internet ingress or egress.

Architecture decisions

Private by default. Requests arrive from the bank’s backends through Amazon Route 53 and AWS Transit Gateway, never from the public internet. Inside the VPC, a Network Load Balancer fronts Kubernetes on Amazon EKS, with images in Amazon ECR, objects in Amazon S3, secrets in AWS Secrets Manager, and keys in AWS KMS. Every AWS service is reached through its own VPC endpoint, so nothing leaves the private network, and the database is reached through a private endpoint into MongoDB Atlas. Production launched across two availability zones and later expanded to three, with disaster recovery on-premises on AWS Outposts.

Production architecture · redrawn from the delivered design
Usersbank customers
Bank backendsPayitIBMBDubai First
Amazon Route 53
AWS Transit Gateway
FAB AWS Cloud · Bahrain region
FAB VPC · private subnets
AWS services · every one reached through a VPC endpoint
EKSECRS3Secrets ManagerKMSCloudFormationLambda
ELB subnetsCore subnetsMSK subnetsSearch subnets Zone 1Zone 2Zone 3
Network Load Balancerone group across the three zones
Core worker node
  • FaceTec Core SDK
  • Dashboard API
  • Dashboard
Core worker node
  • FaceTec Core SDK
  • Dashboard API
  • Dashboard
Core worker node
  • FaceTec Core SDK
  • Dashboard API
  • Dashboard
Amazon MSKbrokers in all three zones
Search worker node
  • FaceTec Search SDK
  • Producer API
  • Consumer API
Search worker node
  • FaceTec Search SDK
  • Producer API
  • Consumer API
Search worker node
  • FaceTec Search SDK
  • Producer API
  • Consumer API
Atlas endpointprivate, no public route
MongoDB Atlas on AWSAtlas VPC · VPC peering · load balancer · three-instance cluster
Disaster recovery · on-premisesAWS Outposts in the bank’s own data centrestandby for the cloud environment

Core and search, scaled separately. Each zone runs two kinds of worker node. Core nodes carry the FaceTec Core SDK with the dashboard and its API. Search nodes carry the Search SDK with the producer and consumer APIs, so face search can grow without touching the liveness path.

Enrol once, reach every node. Each search node keeps its own 3D face database, so an enrolment has to land on all of them. The producer API publishes it to Amazon MSK, which fans it out; a consumer on every node enrols it locally. A Kong ingress controller sends each request down the right path.

Request routing inside a zone
Network Load Balancerfrom the bank backendsKong ingress controllerroutes by path
/sdkFaceTec Core SDKliveness and face match
/searchFaceTec Search SDKlooks up the 3D database
/enrollProducer APIAmazon MSKfans outConsumer APIevery nodeSearch SDKenrols locally

A dashboard and backend of its own. I built the entire dashboard in React and the backend service in Node.js with Express, sitting alongside the FaceTec server and connecting verification to the bank’s application flow.

Inside the bank’s network. The infrastructure had to fit the bank’s existing environment. FAB supplied foundational network resources and approved machine images; I designed and documented how the solution would use them across development and later environments.

Who provided what
FAB provided
  • VPCs and subnets
  • Approved machine images
  • Backend and API gateway
  • Security and network requirements
I delivered
  • Development and production architecture
  • CloudFormation templates and provisioning guide
  • Server-side verification and integration
  • Field-level encryption

Environments from templates. I implemented the infrastructure as code with CloudFormation, with AWS Lambda functions as part of the stack, and wrote the provisioning guide, so environments could be created consistently rather than rebuilt by hand.

Stack parameters · illustrative, not the bank’s configuration
Environment
development | production
VpcId
supplied by the bank
PrivateSubnetIds
supplied by the bank
NodeImageId
bank-approved machine image
VpcEndpoints
one per AWS service
AvailabilityZones
3 in production

Encrypted before storage. I implemented client-side field-level encryption, so sensitive data was protected before it reached database storage.

The customer journey

The Payit flow connected identity-document capture to biometric verification. A customer scanned an Emirates ID and completed a liveness check; the backend then compared the live face with the identity-document image before returning the result to the wallet’s registration journey.

Payit onboarding · verification sequence
  1. 01 · MicroblinkScan Emirates ID
  2. 02 · FaceTecLiveness check
  3. 03 · ServiceMatch face to ID photo
  4. 04 · BankContinue registration
verified → continueno match → bank decidesfailed capture → retry

That placed the verification service inside an existing application and backend workflow. The integration had to handle successful verification, unsuccessful matches, and retry paths while preserving the bank’s control over the wider onboarding process.

What I owned

  1. 01

    Architecture documents for development and production: service selection, environment sizing, private connectivity, and availability.

  2. 02

    Infrastructure as code with CloudFormation and Lambda, with every AWS service behind a VPC endpoint, plus the provisioning guide so environments could be created consistently.

  3. 03

    Technical implementation lead, working with FAB’s teams on environment setup and rollout within the bank’s security and network requirements.

  4. 04

    The dashboard and the backend service, built end to end in React and Node.js with Express.

  5. 05

    Client-side field-level encryption protecting sensitive data before it reached database storage.

  6. 06

    Server-side verification connected to the bank’s application flow, working across customer, engineering, infrastructure, security, and vendor teams.

Where it stands

3availability zones in production, up from two at launch
3bank backends on one shared platform: Payit, IBMB, Dubai First
0direct routes to the external internet

I delivered the initial production rollout and continued supporting releases and operational improvements as the service expanded across the bank’s application network. Production later grew from two availability zones to three, with disaster recovery running on-premises on AWS Outposts.

The engagement brought together the work needed to deliver an enterprise service: a defined user journey, a deployable architecture, repeatable infrastructure, data protection, technical documentation, and coordination across the teams responsible for operating it.

WORKING TOGETHER

Planning a private enterprise deployment? Let’s discuss the requirements.

Let’s talk
NEXT CASE STUDYConnecting digital onboarding to enterprise identity Central Bank of Kuwait