Two Modes, One Engine
1:1 Verification & 1:N Deduplication
Most enterprise journeys need 1:1 — "is this the same person as the one on the document?" National ID programmes also need 1:N — "has this person enrolled before, possibly under a different name?" Both modes ride the same model and ops infrastructure.
1:1 Face Match
Selfie + document photo in, similarity score out. The threshold is a journey-level setting, not a model constant — you can run loose on low-risk wallet sign-ups and strict on V-CIP for personal loans. Output is a structured confidence score, an explanation flag set, and the cropped reference face for audit purposes.
- Configurable threshold per journey / risk band
- Glasses, age progression, partial occlusion handled gracefully
- Returns audit-ready evidence bundle
1:N Deduplication
Used in national ID enrollment to make sure a person doesn't appear in the registry twice. Integrates with an Automated Biometric Identification System (ABIS) when the programme has one, or stands alone for smaller populations. Designed for batch enrollment plus near-real-time check at the registration terminal.
- ABIS integration ready
- Scales to population-level registries
- Adjudication queue for borderline matches
Model Posture
No Third-Party Black Box in the Loop
eMudhra-Trained
No AWS Rekognition, no Azure Face API, no opaque vendor SDK in the verification path. The model is ours, end to end.
Demographic Tuning
Indian and global demographic representation in training. Performance audited per release across skin tone, age, and gender slices.
DPDP / GDPR Aware
Biometric data processing under explicit consent. Templates can be stored as hashes; raw biometrics deleted post-verification on configurable policy.
Mobile SDK
Capture-side enforcement on iOS, Android, and Web — quality checks before the photo leaves the device, reducing bad upload load on the server.
REST API
Server-side calls for back-office review, batch processing, or integration into non-mobile flows.
Evidence Bundle
Every match returns a signed evidence bundle — original captures, model version, score, decision — for audit and dispute resolution.
