RBI V-CIP compliant video KYC with liveness detection, deepfake prevention, and face-Aadhaar matching — all running on-device. Our patent-pending temporal chunking system catches AI-generated spoofs that frame-by-frame analysis misses. Picture-in-Picture live agent verification as prescribed by RBI.
Watch how video KYC completes seamlessly during a borrower's loan origination conversation — no separate app, no redirect, no waiting in queue.
eKYC Demo — Coming Soon
Complete V-CIP flow: consent → edge liveness → document capture → PiP agent verification → face match → done.
As per RBI Master Direction — KYC Direction, 2016 (updated June 2025), V-CIP must be a seamless, live, consent-based audio-visual interaction with a trained official of the regulated entity.
Explicit consent recorded (auditable, tamper-proof). GPS coordinates captured. IP validated within India.
On-device AI checks: sunglasses, mask, hand obstruction, lighting. Anti-spoof before video call begins.
Aadhaar (OTP or offline XML ≤3 days), PAN, address proof captured on camera. OCR extracts data instantly.
Live video with trained RE official. Randomized questions. Agent sees borrower + documents in Picture-in-Picture.
AI face comparison with Aadhaar photo. Confidence score generated. Session recorded with full audit trail.
All liveness and anti-spoofing checks run on-device before the video stream reaches the server. Zero latency, works on low bandwidth, and prevents spoofed video from ever entering the pipeline. Built from our production experience detecting obstructions in another project (eyewear, masks, hands, lighting conditions).
CNN-based detector identifies dark glasses, tinted lenses, reflective eyewear, and any optical obstruction hiding the eye region. Both eyes must be clearly visible and trackable for liveness to pass. Handles prescription glasses vs. sunglasses distinction.
Detects surgical masks, cloth masks, scarves, and partial face coverings. Uses facial landmark mapping — if critical landmarks (nose bridge, mouth corners, chin) are occluded, the session pauses with a clear instruction to remove the obstruction.
Hand-pose estimation detects fingers, palms, or any object partially covering the face. Common in spoof attempts where someone holds a phone displaying a photo in front of their face. Also catches clipboard, paper, or flat-object attacks.
Analyzes ambient lighting to detect: extreme backlighting (silhouette attacks), screen glow patterns from phone/monitor displays, uniform artificial lighting from studio setups, and flash-induced flatness. Natural variance is the hallmark of a live environment.
AI-generated deepfakes fool single-frame detectors. A high-quality face-swap video can pass liveness on any individual frame. But across time, deepfakes leak information that real faces don't. Our patent-pending temporal chunking system exploits this.
The live video stream is segmented into overlapping 0.5-second chunks (~15 frames each). Each chunk is independently analyzed for facial landmarks, skin texture micro-patterns, and pixel-level noise signatures. Individual chunks can look perfectly real — a high-quality deepfake will pass.
Real faces exhibit natural temporal continuity: micro-expressions evolve smoothly, skin texture doesn't reset between frames, blood flow causes subtle color shifts, and lighting interaction changes with head movement. Deepfakes have temporal discontinuities — GAN-generated frames are independently synthesized, creating micro-jitter in texture, inconsistent specular highlights, and unnatural landmark drift between chunks.
A lightweight temporal CNN compares feature vectors across consecutive chunks. Anomalies in: texture continuity (GAN fingerprint flicker), landmark stability (face boundary wobble), frequency spectrum (high-freq noise resets), and lip-audio sync drift (deepfake lip-sync degrades over time). Each anomaly contributes to a composite spoof score.
Why this catches what others miss: Single-frame liveness (used by most providers) checks if this frame looks real. A well-made deepfake passes easily. Our system checks if the transition between frames is physically consistent — something generative models fundamentally struggle with because they synthesize each frame with slight independence. It's the temporal fingerprint of reality.
RBI mandates V-CIP be conducted by a specially trained official of the regulated entity through a live audio-visual interaction. Our Picture-in-Picture interface gives the agent a simultaneous view of the borrower's face and their documents.
The trained RE official sees the borrower's live video as the main view, with their own camera feed in the Picture-in-Picture overlay — exactly like a standard video call. The borrower simultaneously sees the agent on their screen, creating the live two-way interaction RBI mandates. AI-powered OCR and face matching run in the background while the agent focuses on the conversation.
RBI-mandated randomized questions are prompted to the agent from a shuffled pool — ensuring the interaction is live and not pre-recorded. The question sequence varies per session, preventing rehearsal attacks.
Session controls: Pause (brief interruptions don't require restart), flag for review, approve, or reject. If the customer appears prompted or coached off-screen, the agent can terminate and flag the session immediately.
Per RBI KYC Master Direction (updated June 2025), all regulated entities must verify customer identity using Officially Valid Documents (OVDs). Our system captures, verifies, and securely stores all required documentation during the V-CIP session.
Banks: OTP-based Aadhaar e-KYC authentication via UIDAI. NBFCs: Offline Aadhaar verification (XML or Secure QR, must be ≤3 days old). Aadhaar number masked per Section 16. Face photo extracted for AI matching. Consent recorded before submission.
PAN displayed to camera during live V-CIP session. OCR extracts number, name, and DOB. Real-time validation against NSDL/UTIITSL database. Cross-checked with Aadhaar name for consistency. Agent visually confirms the physical card is original.
If current address differs from Aadhaar: utility bill (≤2 months), bank statement, or government letter accepted. DigiLocker documents with valid digital signatures accepted per RBI. Agent captures screen images of all physical OVDs during session.
DocSense-powered OCR runs in real-time during the V-CIP call. Extracts name, DOB, address, document number from displayed documents. Supports Hindi + English on Indian identity documents. Auto-fills the loan application form.
AI checks for manipulation: font inconsistencies, pixel-level editing artifacts, misaligned text baselines, color space anomalies. Compares layout against known templates for Aadhaar, PAN, and Voter ID to flag forged or modified documents.
Integration with Central KYC Records Registry (CKYCR) for uploading records post-verification. DigiLocker-issued e-documents accepted as OVDs. Supports KYC identifier-based verification. All data stored on Indian servers per data localization mandate.
RBI encourages AI/ML-driven facial recognition to validate the customer against their identification document. Our face matching engine compares the live video frame with the Aadhaar photograph.
Best frame from video
128-dim face embedding
Liveness verified on-edge
AI Match
>95% threshold
From e-KYC response
or captured from card
during V-CIP session
Multi-factor verification: Face match alone is never the sole determinant. Our system combines face match confidence + liveness verification + document authenticity + agent visual confirmation + temporal chunk analysis. All five signals must pass for V-CIP approval.
The video KYC market charges per verification, with costs varying by provider and features included. Most charge extra for liveness, deepfake detection, and face matching. We include everything.
| Feature | Traditional In-Person KYC | SaaS Video KYC Providers | Enterprise Video KYC | Augmen eKYC |
|---|---|---|---|---|
| Cost per verification | ₹200–₹500+ | ₹30–₹80 | ₹80–₹150+ | Contact us |
| Liveness detection | Manual only | ✓ Server-side | ✓ Server-side | ✓ Edge AI (on-device) |
| Deepfake detection | ✗ | ₹ Extra add-on | ✓ Frame-based | ✓ Temporal chunking (patent pending) |
| Face–Aadhaar matching | Visual only | ₹ Extra add-on | ✓ Included | ✓ Included |
| Sunglasses/mask detection | Manual | Basic | ✓ | ✓ Edge AI (6 types) |
| Document OCR | ✗ | ₹ Extra add-on | ✓ Included | ✓ DocSense (Hindi + EN) |
| Document tamper detection | ✗ | ✗ | ₹ Extra add-on | ✓ Included |
| PiP agent interface | N/A | ✓ Basic | ✓ Advanced | ✓ Integrated with loan flow |
| Low bandwidth (<100 kbps) | N/A | 500+ kbps needed | Varies | ✓ Optimized for rural India |
| In-conversation (no redirect) | ✗ | ✗ Separate flow | ✗ Separate flow | ✓ Within loan conversation |
| Self-hostable / on-prem | N/A | ✗ Cloud only | Custom pricing | ✓ Default |
| All features included? | Manual process | Add-ons extra | Mostly included | Everything included |
Why we're different: Most SaaS providers charge ₹30–₹80 per basic video call, then add ₹10–₹30 each for liveness, face match, OCR, and deepfake detection. An "all features" verification often costs ₹80–₹150+. Augmen includes every feature — edge liveness, temporal chunking deepfake detection, face-Aadhaar matching, document OCR, tamper detection, and PiP agent interface — at a lower total cost. Contact us for pricing.
Per RBI Master Direction — KYC Direction, 2016 (updated June 12, 2025). V-CIP is treated on par with face-to-face customer identification.
All data transmitted between customer device and hosting is encrypted. Customer consent recording is auditable and alteration-proof. Meets RBI encryption standards for V-CIP infrastructure.
Live GPS coordinates captured with every session. System detects and prevents connections from IP addresses outside India or spoofed IPs. Location data stored with video recording.
V-CIP conducted only by specially trained RE officials. System enforces agent authentication before session access. Question randomization prevents scripted verification.
All V-CIP recordings, documents, and audit logs stored on servers in India. No data leaves Indian borders. Compliant with DPDP Act 2023 and RBI data storage requirements.
Infrastructure undergoes Vulnerability Assessment and Penetration Testing by accredited agencies as prescribed by RBI. Periodic audits. Critical issues resolved before deployment.
All V-CIP accounts undergo concurrent audit. Complete activity logs with date-time stamps maintained. 5-year record retention as per KYC Direction requirements.