High Accuracy KYC Verification System

My goal is to deploy a secure, real-time module that blocks fraudulent sign-ups by matching a user’s live selfie against the photo on their passport, driver’s license, or national ID card. The engine must deliver high-accuracy facial recognition, run liveness checks to prevent spoofing, and extract document data with OCR so I can compare names, dates, and issuing authority automatically.

Scope of work
The system should accept images or video from web and mobile sources, parse MRZ and PDF417 zones where available, and return a clear pass / fail payload to my platform within seconds. A lightweight admin dashboard is needed for manual review, audit trails, and adjustment of match-score thresholds. Please design the backend as REST or GraphQL APIs; Python (FaceNet, InsightFace), Node, or similar proven stacks are all fine as long as they meet the accuracy target and can scale.

Deliverables
• Fully documented API with sample calls
• Liveness detection and face-document matching model achieving ≥99 % true-positive rate at ≤0.01 % false acceptance
• OCR pipeline covering passports, driver’s licenses, and national ID cards
• Admin dashboard with search, review queue, and exportable logs
• Deployment scripts (Docker/Kubernetes) and CI pipeline

Acceptance criteria
A test suite with at least 1 000 mixed document/selfie pairs must run without human intervention and hit the accuracy benchmark above. Latency for a single verification should stay under 5 s on commodity cloud instances.

Include a brief note on how you will handle data encryption, GDPR/CCPA compliance, and future proofing for additional document types so I can evaluate long-term fit.