Executive Summary

Who was the client?  Anand Rathi Group — a leading Indian financial services conglomerate offering wealth management, broking, investment banking, and insurance services with 100+ branches across India.

What challenge did they face?  Manual In-Person Verification (IPV) for new account openings was causing 35% customer drop-offs, costing ₹200+ per verification, and creating 48-72 hour onboarding delays. With growing digital-first customer segments and SEBI’s push for paperless processes, they needed an automated, compliant video verification solution.

Client Background

Business Size

• 100+ branches pan-India
• Multiple sub-groups (wealth management, broking, distribution)
• Thousands of new account applications monthly
• Mix of HNI and retail customers across diverse demographics

Business Model

B2C financial services with B2B2C distribution through sub-brokers and IFAs (Independent Financial Advisors). Revenue dependent on successful account opening velocity and customer activation rates.

Business Challenge

Problems

1. Manual video calls required scheduling, trained officers, and consumed 10-15 minutes per verification
2. High no-show rates for scheduled verification appointments (30-40%)
3. Inconsistent verification quality across different officers
4. No fraud detection beyond visual observation — vulnerable to photo/video replay
5. Paper-based audit trails made compliance reporting time-consuming
6. Regional language customers faced difficulties with English-only verification
7. Branch-dependent process excluded rural and semi-urban customers

Business Impact

• ₹20+ cost per verification (officer time + infrastructure)
• 48-72 hour average time from application to account activation
• 35% drop-off between application submission and verification completion
• Compliance audit preparation took 2+ weeks per cycle
• Fraud incidents averaging 0.8% of new accounts (identity theft, proxy verification)

Project Objectives

1. Reduce verification cost by 70%+
2. Enable instant (sub-60-second) self-service verification
3. Achieve >99% fraud detection accuracy
4. Maintain full SEBI/RBI compliance with digital audit trails
5. Reduce customer drop-off to <15%

Our Approach

Solution Approach

Agile development with 2-week sprints, starting with core verification flow and progressively adding AI layers. User-centric design validated through pilot testing with real customers across demographic segments. Security-first architecture with VAPT compliance built in from day one, not bolted on later.

Solution Delivered — A production-ready, multi-tenant Video KYC platform with:

• Mobile-first responsive web application (no app download required)
• 7-layer AI verification pipeline running asynchronously post-upload
• Officer review dashboard with AI-assisted decision support
• Admin portal with tenant management, analytics, and officer RBAC
• API-first architecture with webhook callbacks for integration
• Automated data lifecycle management (30-day retention)

Architecture / Technology Stack

Layer	Technology
Frontend	React 18, Material UI v9, MediaPipe Vision (WASM), ONNX Runtime Web
Backend	Python FastAPI (async), SQLAlchemy 2.0, PostgreSQL
AI/ML	PyTorch, DeepFace, faster-whisper, IndicWhisper, YOLOWorld, BiSeNet, Silent-Face
Security	RSA encryption (PKCS1v15), JWT + JTI, bcrypt, CSP with nonce
Infrastructure	Docker, nginx reverse proxy, Ubuntu Server
Speech	Web Speech API (client) + faster-whisper (server fallback)

Key Features

1. Guided Self-Service Verification — Step-by-step user journey with real-time face positioning feedback, automatic photo capture on eye-open detection, and multilingual OTP challenge
2. 7-Layer AI Fraud Prevention — Anti-spoofing, liveness challenges, face obstruction detection, headphone/hat detection, forehead visibility check, eye-open validation, and face identity matching
3. Multilingual Speech OTP — Dynamic 4-digit OTP with speech verification in Indian languages, fuzzy matching for accent tolerance, and server-side fallback transcription
4. Intelligent Auto-Verification — 75%+ sessions fully verified by AI without human intervention, with confidence-based routing to officer review queue
5. Officer Review Dashboard — Video playback, AI score breakdown, side-by-side photo comparison, one-click approve/reject with audit notes
6. Multi-Tenant Architecture — Isolated tenant/sub-group structure with independent API credentials, callback URLs, and per-group analytics
7. Enterprise Security — RSA-encrypted passwords, JWT session control with JTI, brute-force lockout, VAPT-compliant headers, automatic data purging
8. Real-Time Analytics — Session volumes, pass rates, failure analysis by sub-group, daily/weekly/monthly trends, officer productivity metrics

Client Benefits

1. Revenue Acceleration — Faster onboarding means faster first-trade/first-investment, improving time-to-revenue
2. Operational Efficiency — Officers handle only 15% of cases (edge cases requiring judgment), freeing them for higher-value customer interactions
3. Pan-India Reach — Customers in tier-2/3 cities complete verification from their phones, eliminating geographic barriers
4. Compliance Confidence — Digital audit trail with video evidence, AI scores, and officer notes readily available for regulatory audits
5. Competitive Positioning — Industry-leading verification experience becomes a differentiator for customer acquisition

Key Learnings

1. Multi-Modal AI > Single Model — No single AI model is sufficient for financial-grade verification. The layered approach (vision + speech + liveness) creates redundancy that catches what individual models miss.
2. Regional Language Support is Non-Negotiable — 40%+ of users chose non-English languages for OTP speech. Without IndicWhisper fallback, these sessions would have failed entirely.
3. Client-Side AI Reduces Latency Dramatically — Running face detection and liveness on-device (MediaPipe WASM) provides instant feedback without server round-trips, critical for the <60-second target.
4. Fallback Mechanisms are Essential — Browser API inconsistencies (Speech API availability, WebGL context limits) require graceful degradation paths. The audio clip fallback saved 15% of sessions that would have failed.
5. Security Must Be Built-In, Not Bolted On — VAPT compliance from design phase (RSA encryption, CSP nonce, JTI tokens) avoided costly refactoring and accelerated security audit clearance.
6. Officer Review Creates a Feedback Loop — Edge cases reviewed by officers provide implicit labels for model improvement, creating a virtuous cycle of increasing automation over time.
7. Mobile-First Means Constraint-First — Designing for mobile GPU limitations (WebGL context budget), intermittent networks (chunked upload), and varying camera quality (adaptive resolution) ensures the solution works for the 80% of users on smartphones.