HyperVerge

HyperVerge is an AI-powered identity verification platform specializing in document OCR (Optical Character Recognition), face matching, liveness detection, and Video In-Person Verification (VIPV). For the broking KYC system, HyperVerge handles the critical visual verification layer — extracting data from identity documents, matching the applicant’s live face against document photos, and conducting SEBI-compliant video KYC sessions. Founded in India and serving major banks, fintechs, and stock brokers, HyperVerge offers both REST APIs and native SDKs across Android, iOS, and Web.

SEBI’s KYC norms require In-Person Verification (IPV) for all new demat and trading accounts. HyperVerge’s liveness-certified face match satisfies the IPV requirement digitally, eliminating the need for physical branch visits. Their iBeta Level 2 certified liveness detection — tested against printed photos, digital screens, video replays, 3D masks, and deepfakes — prevents spoofing attempts and provides a defensible audit trail for regulatory inspections. For the estimated 15-20% of applicants who do not complete DigiLocker-based Aadhaar eKYC, HyperVerge’s VIPV module provides a fully compliant fallback path with tamper-proof recording and geo-tagged activity logs.

This page covers HyperVerge’s API specifications for OCR, face match, liveness detection, and VIPV workflows, including SDK integration for mobile and web, response field mappings, confidence score thresholds, and quality handling strategies. The integration details at the end describe how HyperVerge fits into our broader KYC pipeline alongside DigiLocker, Decentro, and the KYC Admin back-office.

Overview
Face Match API
Liveness Detection
VIPV (Video In-Person Verification)
Document OCR
Non-Individual Entity Documents
SDK Integration
Integration Details
HyperVerge Dashboard
Pricing
Compliance & Certifications
Edge Cases & Failure Handling
Alternatives Comparison
Integration with Our System

1. Overview

About HyperVerge

HyperVerge is an AI-powered identity verification platform founded in India, serving major banks, fintechs, NBFCs, and stock brokers. The company provides document OCR, face match, liveness detection, and video KYC capabilities through REST APIs and native SDKs.

Website: https://hyperverge.co API Docs: https://github.com/hyperverge/kyc-india-rest-api SDK Repos: https://github.com/hyperverge/hyperkyc-android

Our Usage

Capability	Purpose in KYC Flow	Phase
Face Match	Compare DigiLocker Aadhaar photo vs live selfie	Phase 6: Verification
Liveness Detection	Ensure live person (not spoof) during selfie/video	Phase 6: Verification
VIPV	Video In-Person Verification when DigiLocker not used	Phase 6: Verification
Document OCR	Extract fields from PAN, Aadhaar, cheque, passport, DL	Phase 2: Document Capture

Key Metrics

Metric	Value
Face Match Accuracy	~99.8%
OCR Accuracy	~99.8% (with per-field confidence scores)
Liveness Certification	ISO 30107-3 Level 2 (iBeta tested)
Presentation Attack Detection	Printed photos, digital screens, videos, 3D masks, deepfakes
SDK Platforms	Android (5.0+), iOS (11+), Web (JavaScript)
Data Residency	India (Mumbai data center)
Integration Time	1-2 weeks

2. Face Match API

Endpoint

POST https://ind-faceid.hyperverge.co/v1/photo/verifyPair

Authentication

Headers:
  appId: <your_app_id>
  appKey: <your_app_key>

Request

Multipart form-data with two images:

Parameter	Type	Description
`image1`	file (JPEG/PNG) or base64	Reference image (from Aadhaar/PAN photo via DigiLocker)
`image2`	file (JPEG/PNG) or base64	Live selfie (captured via HyperVerge SDK)

POST /v1/photo/verifyPair
Content-Type: multipart/form-data

image1: <aadhaar_photo_file>
image2: <live_selfie_file>

Response

Success (match found):

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "match": "yes",
    "confidence": 92.5,
    "details": {
      "face_detected_ref": true,
      "face_detected_selfie": true,
      "multiple_faces_ref": false,
      "multiple_faces_selfie": false
    }
  }
}

Success (no match):

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "match": "no",
    "confidence": 35.2,
    "details": {
      "face_detected_ref": true,
      "face_detected_selfie": true,
      "multiple_faces_ref": false,
      "multiple_faces_selfie": false
    }
  }
}

Error (face not detected):

{
  "status": "failure",
  "statusCode": "422",
  "error": "Face not detected in image2. Ensure clear, front-facing photo."
}

Score Thresholds and Actions

Score Range	Classification	Onboarding Action
80-100	Strong match	Auto-approve
60-79	Moderate match	Route to manual review queue in KYC Admin
0-59	Weak / no match	Reject or prompt re-capture of selfie

Anti-Spoofing Capabilities

The face match API works in conjunction with liveness detection (Section 3) to detect and reject:

Attack Type	Detection Method
Printed photos	Texture analysis, depth estimation
Digital screens (phone/tablet/laptop)	Moire pattern detection, reflection analysis
Pre-recorded videos	Temporal analysis, lip movement detection
3D masks	Material analysis, skin texture verification
Deepfakes	AI-generated artifact detection, frequency domain analysis

Data Mapping (to Master Dataset)

API Response Field	Master Dataset Field	Section
`result.match`	`face_match_result` (R40)	R: Third-Party Results
`result.confidence`	`face_match_score` (R41)	R: Third-Party Results
`result.details.face_detected_ref`	`face_match_ref_detected` (R42)	R: Third-Party Results
`result.details.face_detected_selfie`	`face_match_selfie_detected` (R43)	R: Third-Party Results

While face match confirms that the person in the selfie matches the document photo, it does not by itself prove that a live human is present. A printed photograph or a video replay of someone’s face could pass a naive face comparison. Liveness detection closes this gap by verifying the physical presence of a real person at the time of capture.

3. Liveness Detection

HyperVerge supports two modes of liveness detection.

3a. Passive Liveness (Recommended for UX)

Attribute	Details
Input	Single selfie image (no gestures required)
Method	AI analyzes texture, depth cues, lighting from a single frame
Processing Time	Sub-second (<500ms typical)
User Experience	Best UX - customer just takes a selfie
Use Case	Default for all onboarding flows

How it works: The customer takes a single selfie through the HyperVerge SDK. The AI model analyzes the image for liveness signals without requiring any user interaction beyond the initial capture. This is transparent to the user.

3b. Active Liveness

Attribute	Details
Input	Video sequence with prompted gestures
Prompts	Head turn left/right, blink, smile, nod
Method	AI verifies real-time response to random prompts
Processing Time	3-5 seconds
User Experience	More friction - requires following on-screen instructions
Use Case	High-value accounts, regulatory requirement, enhanced security

When to use Active Liveness:

Customer flagged for enhanced due diligence (PEP, high-risk)
Passive liveness confidence score falls in borderline range
Regulatory audit requires demonstrable active verification

Liveness API

The liveness check is integrated into the SDK flow. When using the REST API directly:

POST https://ind-faceid.hyperverge.co/v1/photo/verifyLiveness
Content-Type: multipart/form-data
Headers: { "appId": "xxx", "appKey": "xxx" }

Body: { "image": <selfie_file> }

Response:

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "live": "yes",
    "liveness_score": 0.97,
    "attack_type": "none",
    "details": {
      "is_printed_photo": false,
      "is_digital_screen": false,
      "is_video_replay": false,
      "is_mask": false
    }
  }
}

Failure response (spoof detected):

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "live": "no",
    "liveness_score": 0.12,
    "attack_type": "printed_photo",
    "details": {
      "is_printed_photo": true,
      "is_digital_screen": false,
      "is_video_replay": false,
      "is_mask": false
    }
  }
}

Liveness Score Interpretation

Score Range	Classification	Action
0.90-1.00	High confidence live	Auto-approve
0.70-0.89	Moderate confidence	Consider re-capture or manual review
0.00-0.69	Likely spoof	Reject, prompt re-capture in better conditions

Certification

ISO 30107-1 / 30107-3 Level 2 certified
Tested by iBeta Quality Assurance (independent third-party lab)
Level 2 = defends against sophisticated presentation attacks (not just printed photos)
Attacks tested: print (laser/inkjet), digital display (phone/tablet/monitor), 3D mask, video replay

When a customer does not complete DigiLocker-based Aadhaar eKYC, the IPV exemption does not apply, and a Video In-Person Verification session becomes mandatory. VIPV combines face match, liveness, document OCR, and OTP verification into a single recorded video session that satisfies SEBI’s IPV requirements.

4. VIPV (Video In-Person Verification) - SEBI Compliant

When Is VIPV Required?

Scenario	VIPV Required?	Reason
Customer completed Aadhaar eKYC via DigiLocker	NO (Exempted)	IPV exemption per SEBI KYC Master Circular
Customer uploaded Aadhaar Offline XML (not eKYC)	YES	No real-time identity verification done
Customer used alternative address proof (Passport, DL, Voter ID)	YES	eKYC not performed
Customer could not complete DigiLocker flow	YES	Fallback path requires VIPV

Expected usage: ~15-20% of customers (those not using DigiLocker for Aadhaar eKYC).

SEBI VIPV Requirements Checklist

Per SEBI KYC Master Circular (SEBI/HO/MIRSD/MIRSD-SEC-2/P/CIR/2023/168) and Stock Brokers Master Circular (SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/90):

#	Requirement	HyperVerge Implementation
1	Random question generation + response capture	AI generates random questions; speech-to-text captures answers
2	OTP verification (customer reads aloud)	OTP sent to registered mobile; customer reads aloud; auto-recognized via speech-to-text
3	Document display (PAN, Aadhaar, KYC form shown to camera)	SDK prompts customer to hold up each document; OCR runs in real-time
4	Live face match vs document photo	Video frame compared against Aadhaar/PAN photo during session
5	Liveness detection during video	Passive liveness runs continuously during video session
6	Geo-location capture	SDK captures GPS coordinates (lat/long) at session start
7	Tamper-proof recording storage (7-year retention per RBI)	Recording encrypted (AES-256), SHA-256 hash generated, stored in India DC
8	Activity log with timestamps	Every action timestamped: session start, document shown, question asked, OTP read, etc.

VIPV Flow

Step 1:  SDK initiates video session
         -> Camera + microphone permissions requested
         -> Session ID generated
         -> Geo-location captured (lat/long)

Step 2:  Customer shows PAN card to camera
         -> Real-time OCR extracts: PAN number, name, DOB, father's name
         -> Cross-verified against pre-captured PAN data

Step 3:  Customer shows Aadhaar card to camera
         -> Real-time OCR extracts: name, DOB, gender, address
         -> Photo extracted for face match reference
         -> Cross-verified against pre-captured Aadhaar data

Step 4:  Random question displayed on screen
         -> Customer answers verbally
         -> Speech-to-text captures response
         -> Answer logged with timestamp

Step 5:  OTP sent to registered mobile number
         -> Customer reads OTP aloud
         -> Speech-to-text auto-recognizes OTP digits
         -> OTP verified against server-generated value

Step 6:  Live face match
         -> Video frame captured
         -> Compared against Aadhaar/PAN photo
         -> Match score calculated (same thresholds as Section 2)

Step 7:  Liveness check (passive, during video)
         -> AI confirms live person throughout session
         -> No additional gesture required

Step 8:  Geo-location re-confirmed
         -> Final GPS coordinates captured

Step 9:  Session recording finalized
         -> Video encrypted (AES-256)
         -> SHA-256 hash generated for tamper evidence
         -> Recording uploaded to secure storage (India DC)

Step 10: Result generated
         -> PASS / FAIL with detailed scorecard
         -> All individual check results (face match, liveness, OCR, OTP) included
         -> Activity log with all timestamps attached

VIPV Modes

Mode	Description	Pass Rate	Cost	Best For
Agent-Assisted	Human agent conducts VIPV via video call, verifies documents and identity interactively	Higher (~98%)	Rs. 40-50/session	Complex cases, elderly customers, low digital literacy
Fully Automated	AI-driven, no human agent involved, all checks performed by AI	Standard (~95%)	Rs. 30-40/session	Scale, standard onboarding, cost optimization

VIPV API

Initiate Session:

POST https://ind-vkyc.hyperverge.co/v1/session/create
Headers: { "appId": "xxx", "appKey": "xxx" }

{
  "customer_id": "CUST_12345",
  "reference_image": "<base64_aadhaar_photo>",
  "customer_name": "RAKESH KUMAR",
  "customer_mobile": "9876543210",
  "customer_email": "rakesh@example.com",
  "mode": "automated",
  "callback_url": "https://api.broker.in/webhooks/hyperverge/vipv"
}

Session Status Webhook:

{
  "event": "session.completed",
  "session_id": "VIPV_SESSION_ABC123",
  "status": "PASS",
  "result": {
    "face_match": {
      "match": "yes",
      "score": 88.5
    },
    "liveness": {
      "live": "yes",
      "score": 0.96
    },
    "ocr": {
      "pan_verified": true,
      "aadhaar_verified": true
    },
    "otp_verified": true,
    "questions_answered": 2,
    "geo_location": {
      "latitude": 28.6139,
      "longitude": 77.2090
    },
    "recording": {
      "url": "https://storage.hyperverge.co/recordings/VIPV_SESSION_ABC123.mp4",
      "hash_sha256": "a1b2c3d4e5f6...",
      "duration_seconds": 180,
      "size_bytes": 15728640
    },
    "activity_log": [
      {"timestamp": "2026-02-13T10:30:00Z", "action": "session_started"},
      {"timestamp": "2026-02-13T10:30:05Z", "action": "geo_location_captured"},
      {"timestamp": "2026-02-13T10:30:15Z", "action": "pan_shown", "ocr_result": "ABCDE1234F"},
      {"timestamp": "2026-02-13T10:30:45Z", "action": "aadhaar_shown", "ocr_result": "XXXX-XXXX-1234"},
      {"timestamp": "2026-02-13T10:31:00Z", "action": "random_question_asked", "question": "What is your mother's maiden name?"},
      {"timestamp": "2026-02-13T10:31:15Z", "action": "question_answered"},
      {"timestamp": "2026-02-13T10:31:30Z", "action": "otp_sent"},
      {"timestamp": "2026-02-13T10:31:45Z", "action": "otp_verified"},
      {"timestamp": "2026-02-13T10:32:00Z", "action": "face_match_completed", "score": 88.5},
      {"timestamp": "2026-02-13T10:32:05Z", "action": "liveness_confirmed"},
      {"timestamp": "2026-02-13T10:32:10Z", "action": "session_completed", "result": "PASS"}
    ]
  }
}

Recording Storage

Attribute	Details
Encryption	AES-256 at rest, TLS 1.2+ in transit
Tamper Evidence	SHA-256 hash generated at recording completion
Retention	Minimum 7 years (RBI V-CIP requirement)
Storage Location	India (Mumbai data center)
Access	Downloadable via API with authenticated credentials
Format	MP4 (H.264 video, AAC audio)

Data Mapping (to Master Dataset, Section N)

HyperVerge Output	Master Dataset Field	Section
`session_id`	`vipv_session_id` (N09)	N: IPV Details
`activity_log[0].timestamp`	`vipv_start_time` (N10)	N
`activity_log[-1].timestamp`	`vipv_end_time` (N11)	N
`recording.url`	`vipv_video_url` (N14)	N
`recording.hash_sha256`	`vipv_video_hash` (N15)	N
`result.face_match.score`	`vipv_face_match_score` (N17)	N
`result.liveness.score`	`vipv_liveness_score` (N18)	N
`result.activity_log` (questions)	`vipv_random_questions_json` (N19)	N
`result.geo_location`	`vipv_geo_location` (N20)	N
`recording.duration_seconds`	`vipv_duration_seconds` (N16)	N

5. Document OCR

Base URL

Environment	Base URL
Sandbox	`https://test-docs.hyperverge.co/v2.0`
Production	`https://ind-docs.hyperverge.co/v2.0`

Authentication

Headers:
  appId: <your_app_id>
  appKey: <your_app_key>

Supported Documents and Endpoints

Document	Endpoint	Method	Content-Type
PAN Card	`/readPAN`	POST	multipart/form-data
Aadhaar Card	`/readAadhaar`	POST	multipart/form-data
Passport	`/readPassport`	POST	multipart/form-data
Driving License	`/readKYC`	POST	multipart/form-data
Voter ID	`/readVoterID`	POST	multipart/form-data
Cheque / Cancelled Cheque	`/readKYC`	POST	multipart/form-data
Auto-Detect (any KYC doc)	`/readKYC`	POST	multipart/form-data

Input Requirements

Parameter	Details
Field name	`image`
Accepted formats	JPEG, PNG, TIFF, PDF (first page)
Minimum resolution	800px width recommended
Maximum file size	5MB
Color	Color preferred; grayscale accepted with lower accuracy

Request Example (PAN OCR)

POST /v2.0/readPAN
Content-Type: multipart/form-data
Headers: { "appId": "xxx", "appKey": "xxx" }

Body:
  image: <pan_card_image_file>

Response Examples

PAN Card OCR Response

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "details": [
      {
        "type": "pan",
        "fieldsExtracted": {
          "pan_no": { "value": "ABCDE1234F", "conf": 0.99 },
          "name": { "value": "RAKESH KUMAR", "conf": 0.99 },
          "father": { "value": "SURESH KUMAR", "conf": 0.95 },
          "date": { "value": "01/01/1990", "conf": 0.97 },
          "date_of_issue": { "value": "15/06/2015", "conf": 0.85 }
        }
      }
    ]
  }
}

PAN OCR Fields:

Field	Key	Description	Typical Confidence
PAN Number	`pan_no`	10-character alphanumeric PAN	0.97-0.99
Name	`name`	Name as printed on PAN card	0.95-0.99
Father’s Name	`father`	Father’s name on PAN	0.90-0.97
Date of Birth	`date`	DOB in DD/MM/YYYY format	0.93-0.99
Date of Issue	`date_of_issue`	Card issue date (if visible)	0.80-0.90

Aadhaar Card OCR Response

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "details": [
      {
        "type": "aadhaar",
        "fieldsExtracted": {
          "aadhaar_no": { "value": "XXXX-XXXX-1234", "conf": 0.98 },
          "name": { "value": "RAKESH KUMAR", "conf": 0.99 },
          "dob": { "value": "01/01/1990", "conf": 0.97 },
          "gender": { "value": "M", "conf": 0.99 },
          "address": { "value": "123, MG Road, Sector 5, Gurgaon, Haryana - 122001", "conf": 0.88 },
          "photo": { "value": "<base64_encoded_face_photo>", "conf": 0.95 }
        }
      }
    ]
  }
}

Aadhaar OCR Fields:

Field	Key	Description	Typical Confidence
Aadhaar Number	`aadhaar_no`	Masked (last 4 digits visible per UIDAI norms)	0.95-0.99
Name	`name`	Name on Aadhaar	0.95-0.99
Date of Birth	`dob`	DOB in DD/MM/YYYY	0.93-0.98
Gender	`gender`	M / F / T	0.98-0.99
Address	`address`	Full address as printed	0.80-0.92
Photo	`photo`	Extracted face photo (base64)	0.90-0.97

Cheque / Cancelled Cheque OCR Response

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "details": [
      {
        "type": "cheque",
        "fieldsExtracted": {
          "account_no": { "value": "1234567890", "conf": 0.96 },
          "ifsc": { "value": "SBIN0001234", "conf": 0.98 },
          "bank_name": { "value": "State Bank of India", "conf": 0.99 },
          "branch": { "value": "Gurgaon Main Branch", "conf": 0.90 },
          "micr": { "value": "110002345", "conf": 0.94 }
        }
      }
    ]
  }
}

Cheque OCR Fields:

Field	Key	Description	Typical Confidence
Account Number	`account_no`	Bank account number	0.92-0.98
IFSC Code	`ifsc`	11-character IFSC	0.95-0.99
Bank Name	`bank_name`	Name of the bank	0.95-0.99
Branch	`branch`	Branch name	0.85-0.95
MICR Code	`micr`	9-digit MICR code	0.90-0.97

Passport OCR Response

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "details": [
      {
        "type": "passport",
        "fieldsExtracted": {
          "passport_num": { "value": "K1234567", "conf": 0.98 },
          "name": { "value": "RAKESH KUMAR", "conf": 0.99 },
          "dob": { "value": "01/01/1990", "conf": 0.97 },
          "nationality": { "value": "INDIAN", "conf": 0.99 },
          "gender": { "value": "M", "conf": 0.99 },
          "issue_date": { "value": "15/06/2020", "conf": 0.95 },
          "expiry_date": { "value": "14/06/2030", "conf": 0.96 },
          "place_of_birth": { "value": "NEW DELHI", "conf": 0.92 },
          "place_of_issue": { "value": "NEW DELHI", "conf": 0.90 },
          "mrz_line1": { "value": "P<INDKUMAR<<RAKESH<<<<<<<<<<<<<<<<<<<<<<<<", "conf": 0.99 },
          "mrz_line2": { "value": "K12345671IND9001010M3006146<<<<<<<<<<<<<<00", "conf": 0.99 }
        }
      }
    ]
  }
}

Driving License OCR Response

{
  "status": "success",
  "statusCode": "200",
  "result": {
    "details": [
      {
        "type": "dl",
        "fieldsExtracted": {
          "dl_number": { "value": "HR0120190012345", "conf": 0.97 },
          "name": { "value": "RAKESH KUMAR", "conf": 0.98 },
          "dob": { "value": "01/01/1990", "conf": 0.96 },
          "address": { "value": "123, MG Road, Gurgaon, Haryana", "conf": 0.85 },
          "issue_date": { "value": "01/06/2019", "conf": 0.93 },
          "validity": { "value": "31/05/2039", "conf": 0.94 },
          "vehicle_classes": { "value": "LMV, MCWG", "conf": 0.90 }
        }
      }
    ]
  }
}

Auto-Detect Mode (`/readKYC`)

The /readKYC endpoint automatically identifies the document type and extracts relevant fields. Use this when:

Document type is not known in advance
Building a generic document upload flow
Processing mixed document batches

The response includes a type field indicating the detected document type: pan, aadhaar, passport, dl, voter_id, cheque, etc.

Tamper Detection

HyperVerge OCR includes built-in tamper detection that flags signs of document manipulation:

Tamper Type	Detection
Photo pasting (face swapped on document)	Detects paste edges, inconsistent lighting
Text editing (Photoshopped name/DOB)	Font inconsistency, pixel-level artifacts
Cropped/partial documents	Missing expected regions, incomplete fields
Color-copied documents	Color profile analysis, print pattern detection

Tamper detection results appear in the response as:

{
  "tamper_check": {
    "is_tampered": false,
    "tamper_score": 0.05,
    "tamper_details": []
  }
}

Confidence Score Handling

Confidence Range	Quality	Action
0.95-1.00	High	Auto-accept field value
0.80-0.94	Medium	Accept but flag for visual confirmation by customer
0.60-0.79	Low	Present to customer for manual correction
0.00-0.59	Very Low	Prompt re-capture of document

The individual document OCR capabilities described above cover the primary onboarding flow. However, accounts opened on behalf of Hindu Undivided Families (HUFs), partnerships, and corporates require additional document types that do not follow standardized layouts.

6. Non-Individual Entity Documents

HyperVerge’s OCR capabilities extend beyond individual KYC documents to support entity-level verification.

Corporate Documents

Document	OCR Support	Fields Extracted	Accuracy Notes
CIN Certificate	Yes	CIN number, company name, date of incorporation, registered office	Good accuracy on printed certificates
Board Resolution	Partial	Signatory names, company name, date	Limited accuracy on handwritten portions
MOA/AOA	Partial	Company name, objects (first page)	Multi-page documents require page-by-page processing

HUF Documents

Document	OCR Support	Fields Extracted
Karta’s Aadhaar	Yes (standard Aadhaar OCR)	Name, DOB, gender, address, photo
Karta’s PAN	Yes (standard PAN OCR)	PAN number, name, DOB
HUF PAN	Yes (standard PAN OCR)	HUF PAN number, HUF name
HUF Deed	Partial (generic document OCR)	HUF name, Karta name (accuracy varies)

Partnership Documents

Document	OCR Support	Fields Extracted
Partnership Deed	Partial	Firm name, partner names, dates (accuracy varies on scanned legal docs)
Partnership PAN	Yes (standard PAN OCR)	Firm PAN, firm name
Partner’s PAN/Aadhaar	Yes (standard individual OCR)	All standard fields

Face Match for Non-Individuals

For authorized signatories and directors of non-individual entities, the same face match API (Section 2) applies. The reference image is the signatory’s PAN/Aadhaar photo, matched against a live selfie.

7. SDK Integration

Android SDK

Attribute	Details
Repository	https://github.com/hyperverge/hyperkyc-android
Min SDK Version	API 21 (Android 5.0 Lollipop)
SDK Size	~5MB (AAR)
Language	Java / Kotlin compatible
Camera	Uses device camera with quality optimization
Permissions	Camera, Internet, Location (for VIPV)

Gradle Integration:

dependencies {
    implementation 'co.hyperverge:hyperkyc-android:x.y.z'
}

Basic Usage:

val config = HyperKycConfig.Builder()
    .appId("your_app_id")
    .appKey("your_app_key")
    .setWorkflow("face_match_liveness")
    .build()

HyperKyc.start(activity, config, object : HyperKycCallback {
    override fun onSuccess(result: HyperKycResult) {
        // result.faceMatchScore, result.livenessResult, etc.
    }
    override fun onFailure(error: HyperKycError) {
        // Handle error
    }
})

iOS SDK

Attribute	Details
Repository	HyperVerge iOS SDK (contact for access)
Min iOS Version	iOS 11.0+
Language	Swift / Objective-C compatible
SDK Size	~4MB
Permissions	Camera, Internet, Location (for VIPV)

CocoaPods Integration:

pod 'HyperKyc', '~> x.y.z'

Web SDK (JavaScript)

Attribute	Details
Integration	JavaScript SDK loaded via script tag or npm
Browser Support	Chrome 60+, Firefox 60+, Safari 12+, Edge 79+
Requirements	Camera access via getUserMedia API, HTTPS required
Size	~200KB (minified + gzip)

Script Tag Integration:

<script src="https://cdn.hyperverge.co/hyperkyc-web/v1/hyperkyc.min.js"></script>

Basic Usage:

const hyperKyc = new HyperKyc({
    appId: 'your_app_id',
    appKey: 'your_app_key',
    workflow: 'face_match_liveness'
});

hyperKyc.start({
    containerId: 'kyc-container',
    onSuccess: function(result) {
        console.log('Face match score:', result.faceMatchScore);
        console.log('Liveness:', result.livenessResult);
    },
    onFailure: function(error) {
        console.error('Error:', error.code, error.message);
    },
    onCancel: function() {
        console.log('User cancelled');
    }
});

SDK Capabilities

All SDKs handle the following:

Feature	Description
Camera capture	Optimized capture with auto-focus, exposure adjustment
Image quality check	Real-time feedback if image is blurry, dark, or overexposed
Liveness prompts	On-screen instructions for active liveness (if configured)
Document alignment guide	Overlay guide for positioning PAN/Aadhaar in frame
Result callback	Structured result object with scores and status
Error handling	Detailed error codes for camera issues, network errors, etc.
UI customization	Customizable themes, colors, fonts to match broker’s brand identity
Localization	Support for multiple Indian languages in UI text

With the SDK handling client-side capture and quality checks, the backend integration layer is responsible for authentication, API orchestration, error handling, and webhook processing. The following section covers these server-side concerns in detail.

8. Integration Details

Authentication

All API requests require appId and appKey in HTTP headers:

Headers:
  appId: <your_application_id>
  appKey: <your_application_key>
  Content-Type: multipart/form-data (for image APIs)
  Content-Type: application/json (for session APIs)

Credentials are issued per environment (sandbox vs production). Rotate keys periodically (quarterly recommended).

Base URLs

Service	Sandbox	Production
Document OCR	`https://test-docs.hyperverge.co/v2.0`	`https://ind-docs.hyperverge.co/v2.0`
Face Match / Liveness	`https://test-faceid.hyperverge.co/v1`	`https://ind-faceid.hyperverge.co/v1`
VIPV	`https://test-vkyc.hyperverge.co/v1`	`https://ind-vkyc.hyperverge.co/v1`
Dashboard	`https://dashboard.hyperverge.co`	`https://dashboard.hyperverge.co`

Rate Limits

Tier	Rate Limit	Burst
Standard	50 TPS (transactions per second)	100 TPS
Enterprise	100+ TPS (configurable)	200+ TPS

Rate limit headers returned:

X-RateLimit-Limit: 100
X-RateLimit-Remaining: 95
X-RateLimit-Reset: 1707820800

Webhook Support

For async operations (VIPV session results), HyperVerge sends webhooks to the configured callback_url:

Webhook Payload:

{
  "event": "session.completed",
  "session_id": "VIPV_SESSION_ABC123",
  "status": "PASS",
  "timestamp": "2026-02-13T10:32:10Z",
  "result": { ... }
}

Webhook Security:

HMAC-SHA256 signature in X-HyperVerge-Signature header

Verify signature before processing:

signature = HMAC-SHA256(webhook_secret, request_body)

Error Codes

HTTP Code	Error	Description	Action
200	Success	Request processed successfully	Process response
400	Bad Request	Malformed request, missing required fields	Fix input parameters
401	Unauthorized	Invalid or expired appId/appKey	Check credentials, rotate if needed
404	Not Found	Invalid endpoint or session ID	Verify URL and session ID
413	Payload Too Large	Image exceeds maximum file size (5MB)	Compress or resize image
422	Unprocessable Entity	Face not detected, document not readable	Prompt user to re-capture
429	Too Many Requests	Rate limit exceeded	Implement backoff, retry after `Retry-After` header
500	Internal Server Error	HyperVerge server issue	Retry with exponential backoff
503	Service Unavailable	Temporary maintenance or overload	Retry after delay

Retry Strategy

Recommended retry policy for transient errors (429, 500, 503):

Attempt 1: Immediate
Attempt 2: Wait 1 second
Attempt 3: Wait 3 seconds
(Max 2 retries, total 3 attempts)

For 422 errors (face not detected, document unreadable), do NOT retry automatically. Prompt the user to re-capture the image with better conditions.

Timeout Configuration

API	Recommended Timeout
Face Match	10 seconds
Liveness	10 seconds
Document OCR	15 seconds
VIPV Session Create	15 seconds
VIPV Webhook	N/A (async)

9. HyperVerge Dashboard

HyperVerge provides an admin dashboard at https://dashboard.hyperverge.co for monitoring and management.

Dashboard Features

Feature	Description
Real-time Analytics	Total verifications, success/failure rates, average scores by day/week/month
Transaction Search	Search by session ID, customer ID, date range, status
Manual Review Queue	View borderline cases (face match 60-79%) for human decision
Manual Override	Approve or reject borderline verifications with audit trail
Screenshot/Video Viewer	View captured selfies, document images, VIPV recordings inline
Audit Trail	Complete history of every verification attempt with timestamps and reviewer actions
API Usage	Track API call volume, latency percentiles, error rates
Alerts	Configure alerts for high failure rates, API errors, unusual patterns

Manual Review Workflow

1. Verification completes with borderline score (face match 60-79%)
2. Case appears in "Pending Review" queue on dashboard
3. Reviewer sees:
   - Reference image (Aadhaar/PAN photo)
   - Live selfie
   - Face match score
   - Liveness score
   - OCR results (if applicable)
4. Reviewer decides: Approve / Reject / Request Re-capture
5. Decision logged with reviewer ID, timestamp, reason
6. Webhook sent to broker system with final decision

User Roles

Role	Permissions
Admin	Full access: settings, API keys, user management, all data
Reviewer	Manual review queue, approve/reject, view transactions
Analyst	View-only: analytics, reports, transaction search
Auditor	View audit trail, compliance reports (read-only)

10. Pricing

Per-Transaction Pricing (Estimated)

Service	Cost (INR)	Notes
Face Match	Rs. 1-2 per match	Includes anti-spoofing
Passive Liveness	Rs. 1-2 per check	Often bundled with face match
Active Liveness	Rs. 2-3 per check	Higher due to video processing
Document OCR (per document)	Rs. 1-3 per document	Varies by document type
VIPV - Fully Automated	Rs. 30-40 per session	AI-driven, no agent
VIPV - Agent-Assisted	Rs. 40-50 per session	Includes agent time

Additional Costs

Item	Cost	Notes
SDK License	Included	No separate SDK license fee; included in per-transaction pricing
VIPV Recording Storage	Additional	For 7-year retention beyond standard period; negotiate based on volume
Dashboard Access	Included	Standard dashboard included; advanced analytics may be extra
Support	Included (standard)	Dedicated account manager for enterprise plans

Volume Discounts

Monthly Volume	Typical Discount
Up to 10,000 transactions	Standard pricing
10,001 - 50,000	10-15% discount
50,001 - 200,000	15-25% discount
200,000+	Custom enterprise pricing (negotiate)

Estimated Per-Onboarding Cost (HyperVerge components only)

Scenario	Components	Est. Cost
DigiLocker path (no VIPV)	Face Match + Liveness + OCR (2 docs)	Rs. 5-9
Non-DigiLocker path (with VIPV)	Face Match + Liveness + OCR (2 docs) + VIPV	Rs. 35-55

Beyond pricing, the choice of a verification vendor in the Indian broking context is heavily influenced by regulatory compliance posture. HyperVerge’s certification portfolio and data residency commitments are summarized below.

11. Compliance & Certifications

Certifications

Certification	Scope	Status
ISO 27001	Information Security Management System	Certified
SOC 2 Type II	Security, Availability, Processing Integrity controls	Certified
ISO 30107-3 Level 2	Presentation Attack Detection (Liveness)	iBeta certified
GDPR	Data protection (for international operations)	Compliant

Regulatory Compliance

Regulation	Compliance
SEBI VIPV	Fully compliant with SEBI KYC Master Circular requirements for Video IPV
RBI V-CIP	Compliant with RBI Video-based Customer Identification Process guidelines
IRDAI VBIP	Compliant with Insurance Regulatory Authority Video-Based Identification Process
DPDP Act 2023	Data Protection Digital Privacy Act compliance (India)
IT Act 2000	Electronic record and digital signature compliance
PMLA	Prevention of Money Laundering Act - KYC verification support

Data Residency

Aspect	Details
Primary Data Center	Mumbai, India
Data Sovereignty	All Indian customer data processed and stored within India
Backup	Geo-redundant within India
Data Deletion	On request, customer data purged within 30 days (except regulatory retention)

SEBI Audit Requirements

HyperVerge supports the following for SEBI inspection/audit:

Requirement	HyperVerge Support
Transaction audit trail	Complete API logs with request/response (PII masked)
VIPV recording access	Downloadable recordings with hash verification
Verification evidence	Face match scores, liveness results, OCR data exportable
System audit	SOC 2 Type II report available on request
Data retention proof	Logs and recordings retained per configured retention policy

Even with high accuracy rates and robust certifications, real-world onboarding introduces a wide range of edge cases — poor lighting, elderly applicants whose appearance has changed significantly since their Aadhaar photo was taken, laminated documents with glare, and unreliable mobile network connections during VIPV sessions. The following section catalogs these scenarios and prescribes handling strategies.

12. Edge Cases & Failure Handling

Image Quality Issues

Issue	SDK Behavior	Backend Behavior	Resolution
Low light / dark environment	SDK shows real-time brightness warning	Returns 422 or low confidence	Prompt user to move to better-lit area
Camera blur / out of focus	SDK shows focus indicator	Returns 422 or low confidence	Prompt user to hold steady, auto-focus
Glare on document	SDK detects reflection spots	Lower OCR confidence	Prompt user to tilt document, remove glare source
Partial document visibility	SDK shows alignment guide	Incomplete field extraction	Prompt user to fit full document in frame
Very low resolution camera	SDK warns at session start	May not meet minimum 800px	Suggest uploading a scanned copy instead

Face Match Challenges

Challenge	Impact	Handling
Elderly users with significant appearance change from Aadhaar photo (10+ years old)	Lower confidence score (60-75 typical)	Lower auto-approve threshold or route to manual review
Glasses (prescription)	Minimal impact	Handles well; trained on diverse dataset
Sunglasses	Moderate impact	SDK prompts removal; face detection may fail
Masks (COVID/medical)	High impact	SDK prompts removal; face detection will fail
Facial hair change (clean-shaven vs full beard)	Moderate impact (5-10 point drop)	Usually still above 80 threshold; manual review if borderline
Multiple faces in frame	Rejection	SDK detects and prompts single face in frame
Heavy makeup or cosmetic changes	Minor impact	Handles well in most cases
Head coverings (religious)	Minimal impact if face fully visible	SDK accepts if face is visible; turban, hijab (face visible) OK

Document OCR Challenges

Challenge	Impact	Handling
Laminated document with heavy glare	Reduced accuracy (80-90%)	Prompt user to capture at angle to avoid glare
Very old / faded documents	Reduced accuracy (70-85%)	Flag low-confidence fields for manual entry
Handwritten portions on legal documents	Poor accuracy (<70%)	Auto-detect and skip; flag for manual transcription
Non-standard fonts or print quality	Moderate impact	Per-field confidence lets system identify weak fields
Multi-language documents (Hindi + English)	Good support for common Indian languages	OCR trained on bilingual Indian documents
Aadhaar in regional language only	Moderate impact	Supports 12 Indian languages but accuracy varies

VIPV Challenges

Challenge	Impact	Handling
Poor network connectivity	Video quality degrades, session may drop	SDK auto-adjusts bitrate; auto-reconnect with session resume
Background noise	OTP speech recognition may fail	Visual OTP entry fallback available
Customer does not understand instructions	Session failure	Agent-assisted mode recommended for such cases
OTP not received on mobile	Session blocked at OTP step	Allow OTP resend (max 3 attempts); fallback to email OTP
Customer holds document too close/far	OCR during video fails	SDK provides distance guide overlay

System-Level Edge Cases

Scenario	Handling
HyperVerge API down (500/503)	Retry with exponential backoff; if persistent, queue for later; show user-friendly message
Rate limit exceeded (429)	Backoff and retry; alert ops team if sustained
Sandbox/Production credential mismatch	401 error; log and alert; environment-specific config check
Image upload timeout	Retry once; if persistent, suggest smaller file or better network
Webhook delivery failure	HyperVerge retries 3 times over 15 minutes; poll status API as fallback

13. Alternatives Comparison

Feature	HyperVerge	IDfy	Signzy
Face Match Accuracy	~99.8%	~99.5%	~99.0%
Liveness Certification	ISO 30107 Level 2 (iBeta)	ISO 30107 Level 1	Basic liveness
Liveness Mode	Passive + Active	Passive + Active	Active only
VIPV - SEBI Compliant	Full (automated + agent)	Full (automated + agent)	Agent-assisted only
OCR Document Types	7+ Indian doc types	3 task types (OCR/Verify/Both)	14,000+ doc types (180+ countries)
OCR Accuracy	~99.8% with per-field confidence	~99.0%	~98.5%
Tamper Detection	Built-in	Available	Available
SDK	Android / iOS / Web	Android / iOS / Web	Web + offline SDK
Face Match Pricing	Rs. 1-2	Rs. 2-5	Rs. 2-4
OCR Pricing	Rs. 1-3	Rs. 2-5	Rs. 2-4
VIPV Pricing	Rs. 30-50	Rs. 30-50	Rs. 40-60
Integration Time	1-2 weeks	1-2 weeks	2-3 weeks
Certifications	ISO 27001, SOC 2 Type II	ISO 27001	ISO 27001, SOC 2
India Data Residency	Yes (Mumbai)	Yes	Yes
Deepfake Detection	Yes	Limited	No
Low Bandwidth Support	Moderate	Moderate	Best (offline SDK)
Best For	High accuracy at scale, full VIPV	Bundled verification + gov DB check	Low bandwidth, global doc coverage

Why HyperVerge for Our Stack

Highest face match accuracy (~99.8%) reduces manual review queue volume
ISO 30107 Level 2 liveness provides strongest presentation attack defense (regulatory defensible)
Full SEBI VIPV compliance with both automated and agent-assisted modes
Native SDKs for Android, iOS, and Web — covers all our customer channels
Single vendor for face match + liveness + OCR + VIPV reduces integration complexity
Per-field confidence scores on OCR enable intelligent auto-fill vs manual review routing
India data residency (Mumbai DC) meets SEBI/RBI data localization requirements
Tamper detection built into OCR catches document fraud early

14. Integration with Our System

Architecture Overview

KYC App (Frontend - Web/Android/iOS)
    |
    +---> HyperVerge SDK (embedded)
    |       |
    |       +--- Camera capture + quality check
    |       +--- Liveness prompts (if active mode)
    |       +--- Document alignment guide
    |       +--- VIPV session management
    |       |
    |       +--- Returns: images, scores, session data
    |
    v
KYC API Gateway (Backend)
    |
    +---> HyperVerge REST APIs
    |       |
    |       +--- /v1/photo/verifyPair    (Face Match)
    |       +--- /v1/photo/verifyLiveness (Liveness)
    |       +--- /v2.0/readPAN           (PAN OCR)
    |       +--- /v2.0/readAadhaar       (Aadhaar OCR)
    |       +--- /v2.0/readKYC           (Auto-detect OCR)
    |       +--- /v1/session/create      (VIPV)
    |       |
    |       +--- Webhooks -> /webhooks/hyperverge/vipv
    |
    +---> Cross-verification with other vendors
    |
    v
KYC Database
    |
    v
KYC Admin (Back-Office)

Data Flow: Face Match (DigiLocker Path)

Step 1: DigiLocker returns Aadhaar XML
        -> Extract photo from <Pht> element (base64)
        -> Store as reference_image in KYC DB

Step 2: HyperVerge SDK captures live selfie
        -> SDK performs client-side quality check
        -> SDK performs passive liveness check
        -> Returns selfie image + liveness result

Step 3: Backend calls HyperVerge Face Match API
        -> POST /v1/photo/verifyPair
        -> image1 = DigiLocker Aadhaar photo
        -> image2 = live selfie from SDK

Step 4: Process result
        -> If match >= 80% AND liveness = pass -> Auto-approve (store in R40-R43)
        -> If match 60-79% -> Route to KYC Admin manual review queue
        -> If match < 60% -> Prompt re-capture; max 3 attempts then reject

Data Flow: OCR (Non-DigiLocker Path)

Step 1: Customer uploads PAN card image
        -> HyperVerge OCR: POST /v2.0/readPAN
        -> Extract: pan_no, name, father, dob (with confidence scores)
        -> Auto-fill form fields where confidence >= 0.80

Step 2: Customer uploads Aadhaar card image
        -> HyperVerge OCR: POST /v2.0/readAadhaar
        -> Extract: name, dob, gender, address, photo (masked aadhaar_no)
        -> Auto-fill form fields where confidence >= 0.80

Step 3: Cross-verify OCR results
        -> Compare PAN OCR name vs Aadhaar OCR name (fuzzy match)
        -> Compare PAN OCR DOB vs Aadhaar OCR DOB (exact match)
        -> Flag discrepancies for manual review

Step 4: Use extracted Aadhaar photo as reference for Face Match (Step 2 above)

Data Flow: Cheque OCR to Penny Drop

Step 1: Customer uploads cancelled cheque image
        -> HyperVerge OCR: POST /v2.0/readKYC
        -> Extract: account_no, ifsc, bank_name, branch, micr

Step 2: Auto-fill bank details in KYC form
        -> account_number = OCR account_no
        -> ifsc_code = OCR ifsc

Step 3: Feed to Decentro Penny Drop API [V3]
        -> POST /core_banking/money_transfer/validate_account
        -> { account_number: <from OCR>, ifsc: <from OCR>, name: <from PAN> }
        -> Verify account ownership via name match

Step 4: Cross-verify
        -> Decentro beneficiary_name vs PAN name vs Aadhaar name
        -> If all match -> bank details confirmed
        -> If mismatch -> flag for review

Data Flow: VIPV (Non-DigiLocker Path)

Step 1: Customer did not complete DigiLocker Aadhaar eKYC
        -> VIPV required (no IPV exemption)

Step 2: Backend creates VIPV session
        -> POST /v1/session/create
        -> Pass: reference_image (from PAN/Aadhaar OCR photo), customer details

Step 3: SDK launches VIPV session
        -> Customer follows on-screen prompts:
           1. Show PAN -> OCR verifies
           2. Show Aadhaar -> OCR verifies
           3. Answer random question
           4. Read OTP aloud
           5. Face match + liveness (continuous)

Step 4: Session completes
        -> Webhook received at /webhooks/hyperverge/vipv
        -> Store results in Section N fields (N09-N20)
        -> Store recording URL + hash for 7-year retention

Step 5: KYC Admin reviews VIPV result
        -> If PASS: mark IPV as completed
        -> If FAIL: contact customer, offer retry (max 3 attempts)
        -> If borderline: reviewer watches recording, makes decision

Error Handling Strategy

Error Type	Detection	Action	Fallback
API timeout	Timeout after configured threshold	Retry once	Show user-friendly message, allow manual document entry
Face not detected (422)	API returns 422	Do NOT retry	Prompt user to re-capture with SDK guidance
Low face match score (<60%)	Score in response	Allow 2 re-captures	Route to VIPV if repeated failure
API down (500/503)	HTTP status code	Retry with backoff (3 attempts)	Queue for async processing; allow user to continue and verify later
Rate limited (429)	HTTP 429 + Retry-After	Wait and retry	Alert ops team if sustained
Tamper detected on document	`tamper_check.is_tampered = true`	Reject document	Require original document; flag for compliance review
VIPV session dropped	Webhook timeout or session status	Session auto-resume on reconnect	If unrecoverable, create new session

Environment Configuration

sandbox:
  app_id: "HV_SANDBOX_APP_ID"
  app_key: "HV_SANDBOX_APP_KEY"
  ocr_base_url: "https://test-docs.hyperverge.co/v2.0"
  face_base_url: "https://test-faceid.hyperverge.co/v1"
  vipv_base_url: "https://test-vkyc.hyperverge.co/v1"

production:
  app_id: "${HYPERVERGE_APP_ID}"
  app_key: "${HYPERVERGE_APP_KEY}"
  ocr_base_url: "https://ind-docs.hyperverge.co/v2.0"
  face_base_url: "https://ind-faceid.hyperverge.co/v1"
  vipv_base_url: "https://ind-vkyc.hyperverge.co/v1"

common:
  face_match_auto_approve_threshold: 80
  face_match_manual_review_threshold: 60
  liveness_auto_approve_threshold: 0.90
  ocr_auto_accept_confidence: 0.80
  max_selfie_retries: 3
  max_vipv_retries: 3
  api_timeout_ms: 10000
  vipv_timeout_ms: 15000
  retry_max_attempts: 3
  retry_backoff_base_ms: 1000
  webhook_secret: "${HYPERVERGE_WEBHOOK_SECRET}"
  callback_url: "https://api.broker.in/webhooks/hyperverge/vipv"
  vipv_recording_retention_years: 7

This document should be read alongside Vendor Integrations (Sections V7, V8, V9 for HyperVerge context) and Master Dataset for field-level data mappings.

HyperVerge

Table of Contents

1. Overview

About HyperVerge

Our Usage

Key Metrics

2. Face Match API

Endpoint

Authentication

Request

Response

Score Thresholds and Actions

Anti-Spoofing Capabilities

Data Mapping (to Master Dataset)

3. Liveness Detection

3a. Passive Liveness (Recommended for UX)

3b. Active Liveness

Liveness API

Liveness Score Interpretation

Certification

4. VIPV (Video In-Person Verification) - SEBI Compliant

When Is VIPV Required?

SEBI VIPV Requirements Checklist

VIPV Flow

VIPV Modes

VIPV API

Recording Storage

Data Mapping (to Master Dataset, Section N)

5. Document OCR

Base URL

Authentication

Supported Documents and Endpoints

Input Requirements

Request Example (PAN OCR)

Response Examples

PAN Card OCR Response

Aadhaar Card OCR Response

Cheque / Cancelled Cheque OCR Response

Passport OCR Response

Driving License OCR Response

Auto-Detect Mode (/readKYC)

Tamper Detection

Confidence Score Handling

6. Non-Individual Entity Documents

Corporate Documents

HUF Documents

Partnership Documents

Face Match for Non-Individuals

7. SDK Integration

Android SDK

iOS SDK

Web SDK (JavaScript)

SDK Capabilities

8. Integration Details

Authentication

Base URLs

Rate Limits

Webhook Support

Error Codes

Retry Strategy

Timeout Configuration

9. HyperVerge Dashboard

Dashboard Features

Manual Review Workflow

User Roles

10. Pricing

Per-Transaction Pricing (Estimated)

Additional Costs

Volume Discounts

Estimated Per-Onboarding Cost (HyperVerge components only)

11. Compliance & Certifications

Certifications

Regulatory Compliance

Data Residency

SEBI Audit Requirements

12. Edge Cases & Failure Handling

Image Quality Issues

Face Match Challenges

Document OCR Challenges

VIPV Challenges

Auto-Detect Mode (`/readKYC`)