Verifisuite by Vontech

An AI-powered KYC/KYB on-boarding and renewal platform for businesses of all sizes deployed on AWS. Customers use WhatsApp or a web app powered by a team of three agents on Amazon Bedrock Agent Core.

Comment

VerifiSuite by Vontech is an omnichannel, AI-powered KYC onboarding and renewal platform engineered for Nigerian banks and microfinance institutions. The platform automates the entire customer validation lifecycle, from document data extraction and registry validation to biometric liveness checks and expiry-driven re-KYC sweeps. By utilizing a hub-and-spoke multi-agent architecture built on Amazon Bedrock AgentCore, the communication layer is entirely isolated from the reasoning engine, allowing channels, agents, and tools to scale independently.

Inspiration

Financial institutions operating in Nigeria face rigorous compliance mandates from the Central Bank of Nigeria (CBN) and strict data handling regulations under the Nigeria Data Protection Act (NDPA) 2023. Traditional KYC is highly manual and prone to bottlenecks, while digital alternatives often force users to navigate clunky app downloads. We wanted to engineer an automated, enterprise-grade solution that meets users where they already spend their time—on WhatsApp—while utilizing agentic AI to handle the undifferentiated heavy lifting of data extraction, identity validation, and continuous compliance monitoring.

What it does

VerifiSuite handles both initial customer onboarding and ongoing KYC renewals. Customers initiate the process via a WhatsApp Business number or a secure web application. The platform then:

  • Verifies contact details via OTP delivery (SMS/Email).
  • Captures identity documents (NIN, International Passport, Driver's License).
  • Extracts entity data using Amazon Textract and validates it against Nigerian national registries via Youverify, with automated failover to Dojah.
  • Executes biometric liveness and face matching via Amazon Rekognition.
  • Runs a daily automated expiry sweep using Amazon EventBridge Scheduler, proactively messaging customers on WhatsApp to request updated documents before their IDs expire.

How we built it

As detailed in the "AWS CDS Agent Architecture Documentation.pdf", we built a multi-agent system utilizing Amazon Bedrock AgentCore and the Strands Agents framework, deployed entirely via Terraform infrastructure-as-code.

  • Ingress & Orchestration: Web and WhatsApp requests pass through separate ingress paths (buffered by Amazon SQS and API Gateway) into a single Automated Customer Validation Orchestrator Agent.

  • Delegation: The orchestrator detects intent and delegates tasks over the A2A protocol to two specialist sub-agents (KYC Verification and KYC Renewal) using AgentCore Gateway.

  • State & Storage: KYC state and audit trails are managed in Amazon DynamoDB, document storage is handled securely in Amazon S3, and conversational continuity is maintained via AgentCore Memory.

Challenges we ran into

  • Media Streaming Constraints: WhatsApp does not natively support live video streaming, which was required for the Amazon Rekognition Face Liveness checks. We solved this by generating and sending short-lived, secure HTTPS links that briefly bridge the user to an authenticated web portal to complete the liveness challenge.

  • Strict Agent Isolation: Enforcing the architecture topology in code. We had to implement strict boundary discipline to ensure the Renewal Sub-Agent could not bypass the hub and call the Verification Sub-Agent directly.

  • Data Privacy & PII Handling: Adhering strictly to NDPA meant raw identifiers (like BVNs or NINs) could never be stored in plaintext. We implemented comprehensive data minimization, storing only salted fingerprints in DynamoDB and utilizing Bedrock Guardrails to mask PII in model outputs and CloudWatch logs.

Accomplishments that we're proud of

  • Successfully decoupling the AI reasoning engine from the API communication layer, ensuring that bursts in WhatsApp traffic are safely buffered by SQS and never overwhelm the orchestrator.

  • Building an automated daily re-KYC sweep that tracks expiration dates in a DynamoDB Global Secondary Index and executes zero-touch renewal workflows.

  • Implementing a highly resilient identity verification pipeline that utilizes Youverify as the primary provider and seamlessly fails over to Dojah if the registry goes down, eliminating single points of failure.

  • Creating a robust, vendor-agnostic DecisionEngine where the AI assembles the evidence, but strictly deterministic thresholds make the final compliance judgments.

What we learned

Building complex agentic workflows requires uncompromising boundary discipline. AI models are excellent at conversational orchestration and intent routing, but deterministic, highly regulated decisions—like whether a name string matches a national registry record—must be handled by pure, threshold-based code to ensure auditable, predictable outcomes. Additionally, managing asynchronous conversation state across multiple channels requires robust session memory architectures, making AgentCore Memory critical for allowing users to drop off mid-journey and resume perfectly.

What's next for VerifiSuite by Vontech

Our immediate next steps involve expanding the corporate/business KYC intake (KYB). We will activate the Corporate Affairs Commission (CAC) registry endpoints to fully automate corporate onboarding verifications. Following that, we plan to integrate downstream core-banking APIs to automatically provision bank accounts immediately after a KYC verdict passes, and begin preparations for onboarding the solution to the AWS Marketplace as an enterprise B2B offering.

Built With

Share this project:

Updates

Submission history