How DocEnsure Works

01

Upload Document

Client-side validation + format detection

Upload any supported document for verification. The system immediately validates format, size, and resolution before processing begins.

Supported Formats

PDF PNG JPG TIFF BMP

Quality Gate

File size check (max 100MB)
Resolution check (min 640x480)
Format validation
DPI check from EXIF data
Blank page detection
PDF encryption check

Auto-Detection: Document type is automatically detected with a confidence score (e.g., "Salary Slip — 94% confidence").

02

Quality Pre-Check

Server-side validation + quality scoring

After upload, the server performs deeper quality checks. Documents that fail minimum quality standards are blocked before verification begins — preventing wasted processing on unreadable files.

Client-Side Checks

File size validation
Format verification
Resolution check

Server-Side Checks

PDF page count analysis
Text layer detection
DPI extraction from EXIF
Blank page detection
Corruption check

Good

300+ DPI, clear text

Fair

150-299 DPI, readable

Poor

Below 150 DPI, blocked

03

Candidate Identity Detection

OCR extraction + fuzzy matching

OCR (pytesseract + PyPDF2) extracts text from the document. Regex patterns then extract identity fields like name, DOB, PAN, and Aadhaar. The system fuzzy-matches against existing candidates to auto-link documents.

Fields Extracted via Regex

Full name
Date of birth, Gender
PAN (ABCDE1234F)
Aadhaar (XXXX XXXX XXXX)
Passport, DL, Voter ID
Employee ID
Phone, Email

5-Signal Matching

Exact ID match (PAN, Aadhaar)
Fuzzy name (Levenshtein)
DOB match
Filename pattern match
Employee ID cross-ref

Exact Match

ID numbers match perfectly. Auto-linked to existing candidate profile.

Probable Match

Strong similarity detected. User confirms the correct candidate.

Ambiguous Match

Multiple potential matches. User selects from a shortlist.

No Match

No existing candidate found. New candidate profile created automatically.

04

AI Verification — 142 Checks

11 categories running in parallel

The core verification engine runs 142 independent forensic checks across 11 categories simultaneously. Each category specializes in a different aspect of document authenticity.

Visual & Layout 18

Signature Verification 12

Metadata Analysis 18

Image Forensics 16

Content Verification 18

Security Features 14

Database Verification 18

AI/ML Detection 10

Audit & Compliance 10

Document Liveness 5

Biometric Verification 5

142

Total Checks

11

Blended Scoring Formula

                score = 0.6 * avg_confidence + 0.4 * (1 - penalty_score)
              

60% weight: average confidence across all checks
40% weight: inverse of penalty score from failures

Critical Failure Caps

3+ critical failures = max score 35
2 critical failures = max score 50
1 critical failure = max score 65

Risk Levels

Very Low

85-100

Low

65-84

Medium

45-64

High

20-44

Critical

<20

CLEAR

Accept

REVIEW

Manual check

SUSPICIOUS

Don't accept

FRAUDULENT

Reject

06

Report & Candidate Profile

PDF report + auto-enriched profiles + audit trail

A comprehensive PDF report is generated with the verdict banner, category-by-category breakdown, and individual check results. The candidate profile is automatically enriched with data from each verified document.

PDF Report Contents

Verdict banner (Clear / Review / Suspicious / Fraudulent)
Overall authenticity score
Category breakdown with scores
Individual check results (142 items)
Critical findings highlighted
Actionable recommendations

Candidate Profile

Auto-enriched with each new document
All verified documents linked
Verification history timeline
Cross-document consistency checks
Full audit trail with timestamps
Downloadable PDF reports

verification_report_summary.json

{
  "document_id": "doc_8f3a2b1c",
  "candidate": "Rajesh Kumar Sharma",
  "document_type": "salary_slip",
  "verdict": "SUSPICIOUS",
  "score": 38,
  "risk_level": "HIGH",
  "checks_passed": 118,
  "checks_failed": 16,
  "checks_warning": 8,
  "critical_failures": 3,
  "top_findings": [
    "PDF edited with iTextSharp after creation",
    "Gross Pay does not match component sum",
    "7 font families detected (expected 1-2)"
  ],
  "recommendation": "Do not accept. Request original payslip directly from employer.",
  "report_url": "/reports/doc_8f3a2b1c.pdf",
  "verified_at": "2026-03-28T10:16:42Z"
}