What is 9thSense and why it exists
Introduction
9thSense is a Visual Intelligence Platform for teams that need to trust what a document, image, or video is telling them — not just read it.
Reading is the easy part. Extraction is table stakes: any OCR pipeline can pull a name off a PAN card. The hard question in 2026 is whether the PAN card is real, whether the selfie next to it is a live human and not a face-swap, whether the bank statement was issued by a bank or generated by a model this morning, and whether all of it belongs to the same person.
9thSense is built around that question. Every verification runs a layered suite of fraud and integrity checks — deepfake detection, liveness, face likeness, lip-sync integrity, synthetic-document detection, cross-verification — governed by declarative Goal Rules that decide when a case may complete. Extraction feeds the checks; the checks make the decision.
What the platform catches
Deepfakes & face swaps
Face-swapped and AI-generated video in selfies and live KYC sessions, with timestamped markers a reviewer can jump to. See Deepfake Detection.
Presentation attacks
Printed photos, screen replays, and masks — caught by passive, active, and whole-session Liveness checks.
Impersonation
Face Match & Likeness scores the live face against the ID photo, searches watchlists and prior applicants, and a face anchor confirms the same person stays on camera for the entire session.
Voice fraud on calls
Dubbed audio, off-camera coaching, and cloned voices — detected by Lip-Sync & AV Integrity analysis of the session.
Synthetic & tampered documents
AI-generated bank statements, edited salary slips, template forgeries — flagged by document integrity checks: layout and font anomalies, metadata inconsistencies, arithmetic that doesn't hold.
Composite identities
Genuine documents from different people stitched into one application — caught by Cross-Verification of every name, date, address, photo, and statement in the case, and entity resolution across cases.
Goal Rules: policy as configuration
Checks produce verdicts; Goal Rules turn verdicts into decisions. Each agent carries a declarative checklist — deterministic rules (ID checksums, date validity, balance thresholds), confidence thresholds on fraud verdicts, and plain-language custom rules the AI evaluates — and a case cannot complete until every rule passes. Rules are versioned and snapshotted into each case, so the audit trail records not just the decision but the exact policy that made it.
And everything around it
The checks sit inside a complete verification platform:
- Document extraction — 40+ pre-built document types across identity, financial, business, travel, and scene categories; PDFs, images, and scans; custom types from the dashboard. See the Library.
- Agents — pipeline, guided-conversation, and autonomous modes that orchestrate collection, extraction, checks, and decisions. See Concepts.
- Cases — one auditable unit per verification, with immutable execution records, agent snapshots, and goal progress.
- Facial biometrics APIs — 1:1 match and 1:N search across 6 configured databases. See Face Biometrics.
Who it is for
Fintechs and lenders — KYC with real fraud defence: liveness + deepfake + likeness on the applicant, integrity checks on income documents, cross-verification before disbursement.
Insurance — Claims evidence you can trust: tamper and synthesis checks on submitted documents, scene photo analysis, cross-verification of the whole claim file.
Video-KYC and contact-centre compliance — Session-long liveness, face anchoring, and lip-sync integrity on live calls; auditable check evidence with timestamped markers.
Compliance teams — Every check verdict is an immutable, versioned execution record. Reconstruct exactly what was checked, what it found, and which rule made the decision — months later.
Enterprise ready
SOC 2 compliance is covered through Digio's certified infrastructure. All data-at-rest is encrypted; all data-in-transit uses TLS 1.3.
Self-hosted AI is available. Configure allow_cloud=false on any agent and no document data leaves your perimeter.
RBAC is enforced on all API endpoints using scoped keys (read / write / admin).
Idempotency is built in. Pass Idempotency-Key on any write operation for safe retries.
Rate limiting is applied per tenant and configurable per key.
Next steps
Checks Overview
Start with the heart of the platform: what each fraud check catches and how verdicts gate a case. See Checks Overview.
Quickstart
Run your first verification in 5 minutes with the Python SDK or a single cURL call.
Goal Rules
The declarative policy language behind every decision. See Goal Rules.
Concepts
The core primitives: Document Types, Checks, Agents, Cases, and Executions.