Veritas

Inspiration

We kept running into the same reality: third-party onboarding for vendors and clients takes weeks, lives in email/PDF chaos, and every team (security, privacy, fraud, legal) sees a different slice of the truth. Many “AI onboarding” solutions assume you can remove humans overnight. That’s not safe or realistic.

Veritas exists to fix that gap: keep humans in control, while giving them instant, structured, explainable signals instead of detective work.

How we built it

Dynamic Portal: Profile-aware onboarding form that adapts to third-party type, data sensitivity, and region, emitting clean structured data.
Workflow Orchestrator (FastAPI): Validates input, classifies risk profile, and fans out checks to multiple engines in parallel.
Modular Engines:
- Security Baseline (IAM, MFA, encryption, logging, network)
- Privacy / PII detection and masking
- Fraud & Integrity (domain, bank, identity consistency)
- Behavioral anomalies
- Consistency vs documentation
Explainable Scoring: Each engine returns its own score, reasons, and confidence, combined into a clear overall recommendation.
Dashboard: Single view for Approve / Review / Block with evidence, plus audit trails and basic metrics.

Challenges we faced

Designing signals that feel realistic and defensible within a hackathon timeframe.
Keeping humans firmly in the loop while still showing clear automation value.
Making outputs explainable without overwhelming reviewers.
Modeling enterprise-style flows (tickets, approvals, baselines) credibly in a demo environment.

What we learned

Governance-grade AI must be transparent, modular, and reviewable, not a single black-box score.
Encoding policy (baseline controls, routing, thresholds) as logic is as important as any model.
Good reviewer UX (clear reasons, layered detail) is critical for real adoption.
You can meaningfully speed up onboarding without trying to replace human judgment.