Healthcare-Grade Security. No Compromises.
Patient data security is not a feature. It is architecture. RBAC/ABAC, column-level encryption, schema-per-tenant isolation, audit trails, and AI guardrails -- because hospitals cannot afford to learn about security from an incident.
Platform Security
Built in from day one. Not added after the first compliance audit.
RBAC + ABAC Access Control
Permissions per role AND per data attribute. A nurse sees vitals. A billing clerk sees charges. Same patient record, completely different views. No code changes -- all policy.
Column-Level PII Encryption
Names, phone numbers, Aadhaar numbers, addresses -- encrypted at the column level with AES-256. Even a database dump is useless without the decryption keys, which require explicit role authorization.
AD Authentication Integration
Active Directory and LDAP for enterprise SSO. Hospital IT manages their own users. MFA. Configurable session timeouts. We do not own your identity layer -- you do.
Multi-Tenant Isolation
Schema-per-tenant. Each hospital's data lives in its own database schema. No shared tables. No accidental cross-tenant leakage. Not row-level security with a tenant_id column. Actual schema isolation.
Audit Trails
Every data access, every API call, every user action -- logged in partitioned audit tables. Who, what, when, where, why. Tamper-evident and queryable. Not optional.
On-Prem + Cloud Deployment
Your data, your rules. Deploy on hospital premises for full sovereignty or in our managed cloud. Same codebase, same security posture either way. Docker and Kubernetes.
Security Architecture
Four layers. Each one enforces security independently. Compromise one, the others still hold.
LLM Safety Guardrails
Clinical AI is not a chatbot you can afford to get wrong. Every AI interaction passes through multiple safety layers before anything reaches a patient.
Topic Boundaries
Each LLM agent has a defined clinical scope. Ask it something outside that scope, and it declines. Not a prompt instruction -- a hard boundary enforced by the policy engine.
Escalation Triggers
Emergency keywords, suicidal ideation signals, acute symptoms -- immediate escalation to human staff. The AI does not respond to life-threatening situations. A human does.
Grounded Responses
Every AI response pulls from actual patient data in NovaHub. Not from the LLM's training data. Every factual claim traces to a specific record.
Output Validation
Before any AI-generated content reaches a patient: medication names checked against formulary, dosages cross-verified. Hallucinated clinical data does not leave the system.
Prompt Governance
Agent prompts are version-controlled markdown files. Tool definitions validated against schemas. No prompt change goes to production without review, testing, and audit.
Safety Monitoring
Real-time monitoring of AI interactions. Unusual patterns, repeated escalations, boundary-testing behavior -- all flagged. Clinical safety officers get their own dashboard.
Security Deep Dives
For teams that need to see the implementation, not just the checklist.
Multi-Tenant Isolation: Schema-Per-Tenant
Each hospital gets its own database schema. Not row-level security with a tenant_id column -- complete schema isolation. Hospital A's tables literally do not exist in Hospital B's namespace. Connection pooling is tenant-aware: every database connection scopes to a single schema. Even if application code had a bug, the database layer prevents cross-tenant queries. That is the point -- security should not depend on developers remembering to add a WHERE clause.
Column-Level PII Encryption
Patient Personally Identifiable Information is encrypted at the column level using AES-256 column-level encryption. This means even a database dump or backup file does not expose patient names, phone numbers, or identification numbers. Decryption keys are managed separately from the database and require explicit role authorization.
Audit Trails That Actually Work
Every action, every API call, every data access -- logged in partitioned audit tables. Time-partitioned for fast queries and sane retention management. Each entry: user identity, action type, resource, timestamp, source IP, session ID, tenant context. Append-only and tamper-evident. When a regulator asks "who accessed this patient's data?" -- the answer takes seconds, not days.
Deployment Options
Same codebase, same security posture whether you deploy on hospital premises or in the cloud. Your IT team decides based on data sovereignty requirements and infrastructure readiness. We do not push either option -- both work.
On-Premises
- ✓Data never leaves hospital network
- ✓Docker/Kubernetes deployment
- ✓Hospital IT manages infrastructure
- ✓Air-gapped option available
Managed Cloud
- ✓THB-managed infrastructure
- ✓India-region data centers
- ✓Automatic updates and patching
- ✓99.9% uptime SLA
Compliance Alignment
DPDPA (India)
Consent management, data minimization, purpose limitation, and data subject rights. Not a compliance checklist -- built into how data flows through the platform.
HIPAA Standards
Access controls, audit controls, integrity controls, transmission security. We meet the HIPAA Security Rule technical safeguards.
ABDM Integration
Compatible with Ayushman Bharat Digital Mission for health data exchange. FHIR-based interoperability baked into the data model.
ISO 27001 Practices
Risk assessment, incident response, continuous improvement. Aligned with ISO 27001 -- the framework, not just the badge.
Security Assessment
Request a security assessment or talk directly to our infrastructure team about your hospital's specific requirements.