Security & Compliance

Security, Compliance, and Data Protection Architecture

This document covers the security controls, encryption standards, deployment isolation models, regulatory compliance posture, responsible AI governance framework, and audit trail architecture implemented within the Ooumph sovereign AI platform. It is intended for procurement officers, security evaluators, and compliance teams conducting due diligence.

Data Handling

Data Handling Philosophy

The foundational architectural principle governing all data interactions within the Ooumph platform.

The Model Comes to the Data

Conventional AI architectures require institutions to export sensitive data to centralised cloud environments for processing. This creates data sovereignty risks, regulatory exposure, and permanent loss of institutional control over sensitive information.

Ooumph inverts this architecture. AI models are deployed directly into the institution's own secure computing environment. All inference, training, and data processing operations execute within the institutional perimeter. No raw data is transmitted to Ooumph infrastructure, third-party cloud services, or any external system.

The constraint is physical. In on-premise and air-gapped configurations there is no code path by which the platform can extract institutional data, which is a stronger guarantee than any clause in a contract can offer.

Encryption

Encryption Standards

Specifications for data protection at rest and in transit across all deployment configurations.

Encryption at Rest

AES-256 Standard

AES-256 encryption applied to all stored data, model weights, and inference outputs
Hardware Security Module (HSM) integration for cryptographic key management
Key rotation enforced on a configurable schedule (default: 90-day cycle)
Encrypted backups with separate key hierarchy and geo-redundancy controls
Full disk encryption on all deployment nodes using LUKS or BitLocker depending on host OS
Database-level encryption for structured data stores (Transparent Data Encryption)

Encryption in Transit

TLS 1.3 Enforced

TLS 1.3 enforced on all API endpoints, inter-service communication, and client connections
Perfect Forward Secrecy (PFS) enabled across all transport channels
Certificate pinning available for high-security deployment configurations
Mutual TLS (mTLS) for service-to-service authentication within the platform mesh
HTTP Strict Transport Security (HSTS) headers with minimum 1-year max-age directive
No fallback to TLS 1.2 or earlier protocols permitted in production environments

Deployment

Deployment Isolation Architecture

The Ooumph platform supports multiple deployment topologies. Each provides a distinct level of network and physical isolation to match institutional security requirements.

On-Premise Deployment

Complete platform installation within the institution's own data centre. All compute, storage, and networking resources remain under full institutional control. No external network dependencies required for core inference operations.

All AI inference and training executed within institutional perimeter
No outbound data transmission to Ooumph or any third-party infrastructure
Full administrative control retained by institutional IT team
Compatible with existing government data centre infrastructure standards

Air-Gapped Deployment

Zero network connectivity to external systems. Designed for classified, defence-grade, and high-security environments where physical isolation is a mandatory requirement.

Complete physical and logical network isolation from public internet
Offline model updates delivered via secure, authenticated media transfer
Tamper-evident packaging for all software distribution artefacts
Suitable for environments governed by national security classification protocols

Hybrid Deployment

Sensitive workloads execute on-premise while scalable, non-sensitive workloads operate within the institution's approved cloud environment. Data classification policies govern workload routing.

Policy-driven workload routing based on data sensitivity classification
Encrypted data channels between on-premise and cloud components
Institutional cloud environment (AWS GovCloud, Azure Government, or equivalent)
Unified management interface across both deployment zones

Isolation Boundary Summary

Attribute	On-Premise	Air-Gapped	Hybrid
Network Isolation	Institutional network only	Complete physical isolation	Split: on-premise + approved cloud
Data Egress	None	None	Non-sensitive only, policy-governed
Update Mechanism	Controlled internal channel	Authenticated offline media	Controlled internal + cloud sync
Suitable For	State and central government	Classified and defence-grade	Enterprises and larger departments

Certifications

Certification and Empanelment Status

Current certifications, empanelments, and certification roadmap. Dates and statuses reflect the most recent update to this document.

ISO 27001:2022

In Progress

Information Security Management System certification. Gap analysis completed. Remediation in progress. Certification audit targeted for Q3 2026.

SOC 2 Type II

Planned

Service Organization Control assessment for security, availability, and confidentiality. Engagement with accredited audit firm scheduled for Q4 2026.

UP DESCO Empanelment

Active

Current empanelment with Uttar Pradesh Digital Economy and Startup Corporation for government IT services delivery.

RailTel Approved Business Partner

Active

Approved business partner status with RailTel Corporation of India Limited (Navratna PSU under Ministry of Railways).

NICSI Engagement

Active

National Informatics Centre Services Inc. (NICSI) engagement for government of India IT project delivery.

Regulatory

Digital Personal Data Protection Act Compliance

Detailed statement on the Ooumph platform's compliance posture with respect to the Digital Personal Data Protection Act, 2023 (DPDP Act).

The Ooumph platform is architected to support institutional compliance with the Digital Personal Data Protection Act, 2023. The following capabilities are built into the platform at the infrastructure level and are available to all deploying institutions as configurable compliance controls.

All personal data processing occurs within Indian territorial boundaries unless explicitly authorised otherwise by the data fiduciary
Data Principal rights (access, correction, erasure, grievance redressal) are architecturally supported through platform APIs
Consent management framework provides granular, purpose-specific consent collection, storage, and withdrawal mechanisms
Data retention policies are configurable per deployment, with automated deletion workflows upon retention period expiry
Data breach notification infrastructure supports the mandatory 72-hour reporting timeline to the Data Protection Board
Data Protection Impact Assessment (DPIA) documentation is maintained and available for institutional review upon request
Processing purpose limitation is enforced at the inference engine level. AI models cannot access data beyond their declared processing purpose
Significant Data Fiduciary obligations (where applicable) are addressed through enhanced audit controls and periodic compliance reporting

Compliance Responsibility Model

Ooumph provides the technical infrastructure and configurable controls necessary to achieve DPDP Act compliance. The deploying institution retains responsibility for defining processing purposes, obtaining Data Principal consent, configuring retention policies, and designating a Data Protection Officer where required by statute. Ooumph provides implementation support, documentation, and ongoing compliance advisory as part of the deployment engagement.

Governance

Responsible AI Governance Framework

The Ooumph platform incorporates a responsible AI governance framework across four areas: explainability, human oversight, bias monitoring, and decision auditability.

Explainability

Every AI-generated output includes a structured reasoning trace documenting the inference path
Decision factors are ranked and presented in human-readable format for officer review
Model confidence scores are reported alongside all outputs to support informed human judgement
Explainability documentation is generated in formats suitable for RTI responses and legal proceedings

Human Oversight

Human-in-the-loop governance is mandatory for all consequential decisions. AI augments human decision-making; it never replaces accountability
Configurable escalation thresholds route low-confidence outputs to designated human reviewers
Override capability is preserved at every decision point. No AI output is auto-executed without human authorisation
Audit records capture both AI recommendations and final human decisions for complete accountability

Bias Monitoring

Continuous statistical monitoring for demographic and group-level output disparities
Pre-deployment fairness testing conducted against defined protected characteristics
Periodic bias audit reports generated and available for institutional compliance review
Model retraining triggers are activated when bias metrics exceed defined thresholds

Decision Auditability

Complete, immutable record of every input, output, model version, and configuration state for each inference
Audit records are cryptographically signed and tamper-evident
Retention periods for audit records are configurable to meet institutional and statutory requirements
Export capabilities support regulatory examination, internal audit, and judicial discovery processes

Audit

Blockchain-Based Audit Trail Architecture

Technical overview of the immutable audit trail system that underpins legal defensibility, RTI compliance, and institutional accountability for all AI-assisted decisions.

Architecture Overview

Every inference, query, and AI-assisted decision processed through the Ooumph platform generates an immutable audit entry on the blockchain audit layer. This layer operates as a tamper-evident, append-only ledger that provides cryptographic proof of decision integrity and a complete, verifiable chain of custody for every AI-generated output.

Every AI interaction generates an immutable, chronologically ordered audit entry on the blockchain layer
Cryptographic hash chaining ensures tamper detection. Any modification to historical records is computationally detectable
Each audit entry includes: timestamp, model version, input hash, output hash, confidence score, and operator identifier
Audit records are structured for direct compatibility with RTI Act Section 4 disclosure requirements
Legal defensibility is supported through verifiable reasoning chains that reconstruct the complete decision pathway
Institutional administrators can generate compliance reports covering any specified time range or decision category
Third-party audit access is supported through read-only, role-based access controls without exposing operational data
Retention policies comply with institutional records management requirements and applicable statutory mandates

Audit Record Schema

Each audit entry contains the following fields, structured for machine-readable export and human review:

Field	Description
Timestamp	UTC timestamp with millisecond precision
Transaction ID	Unique, cryptographically generated identifier
Model Version	Exact version identifier of the model used for inference
Input Hash	SHA-256 hash of the input data (raw input is not stored in the audit layer)
Output Hash	SHA-256 hash of the generated output
Confidence Score	Model confidence metric for the generated output
Operator ID	Authenticated identity of the officer or system initiating the query
Policy Context	Applicable policy rules active at inference time
Previous Hash	Cryptographic link to the preceding audit entry (chain integrity)

Documentation

Request the Security & Compliance Brief

A detailed security brief is available for procurement officers and security evaluators conducting formal due diligence. The document includes expanded technical specifications, architecture diagrams, compliance matrices, and deployment reference configurations.

Request Security Brief (PDF)Schedule Security Architecture Review

The security brief is provided upon verification of institutional affiliation. Submit a request through the briefing form with your official institutional email address.