Video KYC Platform Case Study | AI Identity Verification

The challenge

A leading VideoKYC platform - trusted by major private banks and insurance providers across India - faced critical performance and scalability challenges as customer volumes grew rapidly. The platform had to undergo fundamental re-architecture across four dimensions:

⚠️ No AI-driven identity verification. The platform relied on manual agent review with no automated face matching, document validation, or anomaly detection. This made fraud detection slow and inconsistent - agents could miss subtle manipulation of ID documents, and the growing session volume was making full manual review impossible to sustain.

⚠️ Slow customer onboarding causing drop-offs. KYC sessions were taking significantly longer than acceptable, causing customers to abandon the process midway. Each drop-off represented both a lost customer and a compliance gap - accounts were opening without completed KYC in some cases.

⚠️ Performance bottlenecks in core services. Under concurrent session loads - particularly during banking business hours when dozens of agents handled sessions simultaneously - core services experienced response time degradation. Database query latency spiked, the session management service became a single point of contention, and authentication flows slowed under load.

⚠️ Limited scalability blocking growth. The platform's monolithic session handling could not horizontally scale to meet the growing KYC volumes driven by the bank's customer acquisition campaigns. New bank clients were onboarded to the platform but the infrastructure couldn't keep pace - SLA commitments were at risk.

Re-engineered KYC session flow

AI-Enhanced VideoKYC Pipeline - After OKRUTI Re-engineering

📱

Session Init

Redis session cache, JWT auth, instant agent routing

📄

Doc Capture

Live doc validation, AI quality check, liveness detection

🤖

AI Verification

Face match, OCR extraction, anomaly scoring in real-time

✅

Compliance Close

RBI audit trail, encrypted storage, instant result

Our solution

OKRUTI re-engineered the platform using a scalable microservices architecture - and crucially, integrated AI-powered face matching, document validation, and anomaly detection into the live KYC session flow. Every performance bottleneck was profiled and resolved at the architecture level, not patched at the application layer.

→ AI-Powered Face Matching & Document Validation: Integrated ML models that run during live VideoKYC sessions - comparing the customer's live face to the ID document photograph, validating document authenticity, and detecting manipulation. Results are returned in under 2 seconds, allowing the session to flow without agent intervention for standard cases.

→ Redis + GridGain Smart Caching: Deployed a two-tier caching architecture. Redis handles hot session data - agent states, customer context, document metadata - with sub-millisecond retrieval. GridGain in-memory computing handles reference data (bank configurations, compliance rules, product eligibility matrices) that was previously causing repeated database round trips across every session.

→ Database Query Tuning: Conducted full query profiling across all high-frequency KYC operations. Rewrote N+1 query patterns, added targeted composite indexes on session lookup paths, and optimised the session close transaction - which previously locked multiple tables simultaneously - into a non-blocking, phased commit sequence.

→ Microservices with Elastic Scaling: Decomposed the session monolith into independent services - Session Manager, Identity Verifier, Document Processor, Compliance Recorder, Agent Router. Each scales independently based on demand. During peak onboarding windows, only the Identity Verifier and Session Manager services need additional instances, not the entire platform.

→ Workflow Decoupling for Fault Isolation: Separated the compliance recording workflow from the live session flow using async event publishing. A failure in the audit trail recording service no longer blocks or degrades the session experience - the session closes cleanly and the compliance event is reliably processed from the queue.

→ Java & Spring Boot Backend Optimisation: Rebuilt core service internals with virtual thread pools for I/O-heavy operations, eliminated blocking patterns in the agent notification layer, and moved to reactive request handling for the session initiation API - the highest-volume endpoint on the platform.

Results achieved

60%

Faster KYC onboarding - session completion time reduced from average 8 minutes to under 3.5 minutes

10×

Concurrent session capacity - platform now handles 10× peak concurrent sessions without performance degradation

Automated fraud detection - face matching and document anomaly detection running live in every session

↑

Completion rate & NPS - faster sessions drive significantly higher completion rates and customer satisfaction scores

The platform went from struggling under concurrent load to handling our entire peak onboarding window without a single SLA breach. The AI verification reduced our fraud review queue by over 70% - agents now only review genuinely ambiguous cases.

- CTO, Leading VideoKYC Platform Provider

Scaling a KYC or onboarding platform?

Whether you're building a new VideoKYC system or resolving performance bottlenecks in an existing one, our team can help with architecture, AI integration, and compliance.

Book Free Consultation → All fintech solutions

Scaling Video KYC for Major
Banks & Insurance Companies

Scaling a KYC or onboarding platform?

Let's talk about your platform.

Scaling Video KYC for MajorBanks & Insurance Companies

Scaling a KYC or onboarding platform?

Related case studies

Let's talk about your platform.

Scaling Video KYC for Major
Banks & Insurance Companies