Architecture and Components

🏗️ KYC Architecture and Components

📊 KYC Architecture Diagram


🏛️ KYC Architecture and Components

The KYC (Know Your Customer) system is a modular platform for automated identity verification based on uploaded document images, selfies, and other sources. The system architecture flexibly adapts to various business scenarios and requirements — from banking and fintech to marketplaces and government services.

Supported scenarios:

  • Document — the user uploads a photo of a passport, ID card, or other identification document.

  • Selfie with document — verifying the match between the user's face and the document photo.

  • Selfie — verification using a single face image.

  • Liveness check — confirming that a live person is in front of the camera, not a photo or forgery.

  • Combined scenarios — e.g., document + selfie with document + liveness.

Component purpose:

Each KYC component is responsible for a specific stage of processing and verification:

  • Document image processing: detection, alignment, classification, and segmentation of the document into fields (photo, name, dates, etc.).

  • Text recognition (OCR): extracting data and validating checksums.

  • Face image analysis: comparing the face to the document photo, estimating age and gender.

  • Anti‑fraud mechanisms: detecting photocopies, screenshots, Photoshop edits, screen captures, physical wear.

  • Liveness check: detecting whether a living person is in front of the camera.

  • External database integrations: checking the document and user against police, bailiff, and blacklist databases.

  • Decision Engine: the final stage, where all results are aggregated and a final decision is formed: “Approved”, “Rejected”, or “Manual review required”.

Thus, the KYC architecture is built around a set of independent, scalable, and interconnected modules that allow it to quickly adapt to any verification scenario.


🗂️ Architecture and Module Interaction

Text diagram of the typical DOCUMENT + SELFIE WITH DOCUMENT workflow

chevron-rightText diagram of the typical DOCUMENT + SELFIE WITH DOCUMENT workflowhashtag

1. Document Detector

Responsible for detecting the presence of a document in the image, validating quality, and preparing it for further processing.

Functions:

  • File format validation (JPEG, PNG)

  • Image size and resolution check

  • Lighting and shadow check

  • Document alignment and normalization


2. Selfie with Document Detector

Responsible for verifying the selfie with document, matching identity, and detecting fraud attempts.

Functions:

  • Face Matching between selfie and document photo

  • Age and gender consistency check

  • Detection of screen images (photo of a monitor/phone)

  • Detection of graphic editor traces

  • Detection of photocopies/prints

  • Document detection in selfie and document type identification


3. Liveness Check

Responsible for confirming that a live human is in front of the camera. The module analyzes face behavior, skin texture, micro‑movements, and signs of real presence.

Functions:

  • Analysis of image sequences or video for liveness signals

  • Rejection of photos or screen captures

  • Deepfake detection — identifying manipulated or AI‑generated face images

  • Detection of fully synthetic images (GAN/Stable Diffusion/FaceApp etc.)


4. External Database Verification

Responsible for checking the user against official registries, government databases, financial datasets, and sanctions lists.

Functions include:

  • Document & identity checks

    • Passport validity check

    • Driver license verification

    • Passport validity checks (number verification, lost/stolen database)

  • Debts & financial checks

    • Court Enforcement Office debt search

    • IRS/ HMRC tax liabilities

    • Self-employment tax status (IRS Schedule SE / HMRC Self Assessment)

    • Credit bureau score (Equifax / Experian / TransUnion)

    • Short-term loan default risk score

    • Mortgage and lien records search

    • Phone number activity & carrier history

    • Name-to-phone reverse lookup validation

  • Legal and judicial checks

    • U.S. Bankruptcy Register / UK Insolvency Register

    • Commercial court cases (UK: High Court / US: Commercial Division)

    • Civil and criminal court cases

    • Register of company directors (Companies House / SEC records)

    • Corporate affiliation records (SEC/Companies House)

    • Wanted lists: FBI / U.S. Marshals Service / HM Prison & Probation Service

  • Identity data checks

    • Social Security Number (SSN) ↔ full name validation

    • Taxpayer Identification Number (TIN) lookup by name & ID

    • Email ↔ Full name correspondence check

    • Phone number ownership check

  • Sanctions & restrictions

    • Terrorist lists

    • Sanctions lists (EU, UK, OFAC, UN)

    • Frozen business accounts (IRS/OFAC compliance)

    • Foreign Agents Registration Act (FARA) register

    • Child support debtors (Federal Office of Child Support Enforcement)


5. Document Classification & Segmentation

Responsible for determining the document type and extracting important regions.

Functions:

  • Document type classification (passport, driver license, ID card, etc.)

  • Segmentation into fields (photo, text areas, stamps, signatures)


6. OCR & Field Verification

Recognizes the text and validates authenticity.

Functions:

  • OCR text extraction

  • Detection of tampering (inserted/replaced fields)

  • MRZ checksum verification


7. Document Antifraud Checks

Responsible for identifying attempts to forge or alter the document.

Functions:

  • Detection of screen images

  • Detection of photocopies/printed documents

  • Graphic editor manipulation detection

  • Document wear/damage detection

  • Logical consistency validation (dates, numbers, series)


8. Decision Engine

Aggregates all results and produces the final KYC decision.

Functions:

  • Collecting outputs from all modules

  • Producing final status: “Approved”, “Rejected”, “Manual review”

  • Returning error codes and diagnostic descriptions


Last updated