Deployment and security

Start with one scenario and let value compound

A NOVARK rollout has five phases: scenario review, environment setup, PoC validation, go-live and expansion, and each theme can be adopted on its own. NOVARK supports full on-premises deployment, with computing resources planned around usage scale and model needs, and every theme shares one set of access, traceability and data controls.

Plan my rollout FAQ

TIMELINE / 01

Rollout process

Every theme follows the same process. After the first theme, later themes reuse the existing environment, permissions and governance settings instead of starting over.

Phase	Main work	Output
1. Scenario review	Interview business units, choose the first theme and use case, and inventory data sources, integrations and access needs	Scope, data list, success metrics
2. Environment setup	Install on-premises or private cloud, connect accounts and permissions, import knowledge and data	Working test environment
3. PoC validation	Run on real data, build a test question set and accept against the agreed metrics	PoC report and expansion recommendation
4. Go-live	User training, manuals, cut-over and early-stage support	Production environment and handover
5. Expansion	Reuse the same environment and governance to add the next theme or more units	Expanded scope and new metrics

Scope and pace are planned during the scenario review around data conditions, integrations and hardware. A PoC can start on existing equipment or a test environment.

HARDWARE / 02

On-premises deployment guidance

NOVARK supports full on-premises deployment. Computing needs depend mainly on the number of users, the language model chosen and whether OCR and speech recognition are enabled. These tiers show the direction; exact configuration is recommended after the scenario review.

Tier	Scale	Computing resources	Other configuration
Starter	PoC or a single-department pilot	One server with a data-center GPU	Standard enterprise server specifications
Standard	Several departments at once	Multiple data-center GPUs for larger language models	More memory and storage, with a fast internal network
Enterprise	Company or group-wide, several themes	Multiple GPU servers split by inference, OCR and speech, with redundancy	Shared storage, backup and monitoring

Application services and vector search run on standard CPU servers and do not need GPUs.
Speech recognition for customer service and OCR for large volumes of scanned documents use extra GPU capacity and should be included in sizing.
Linux servers with containerised deployment are recommended, which keeps updates and expansion simple.
With a cloud model, only application and data services run on-premises, and prompts are sent to that model provider.

SECURITY / 03

Security and compliance

NOVARK puts security controls at the platform level, so every theme shares one set of controls instead of configuring each theme separately.

Data stays in-house

Full on-premises deployment keeps data, knowledge and records in your own environment. With a cloud model, prompts are sent to that provider, so you can decide its scope by data sensitivity.

Role-based access

AI features and knowledge-base access are set per role and per person. Revocation takes effect immediately, including in existing conversations.

Complete audit trail

One trace ID links user actions, model calls and execution records, so you can see who used what, which model ran and what happened, supporting audits and accountability.

Approved tools only

External tools must be approved before AI can use them, are checked for changes before each use, and can be limited by role, scope or read-only mode.

Personal data protection

Personal-data masking on knowledge import, conversation retention periods and record deletion help meet Taiwan's Personal Data Protection Act.

Prompt safety

Prompt checks and sanitisation of imported content reduce prompt injection and wrong answers, and the AI says so when evidence is insufficient.

Platform controls map to common requirements of ISO/IEC 27001 information security, ISO/IEC 42001 AI management systems and Taiwan's Personal Data Protection Act. NOVARK supports your security audits and vendor assessments with architecture descriptions and control documentation.

SCENARIOS / 04

Example rollout scenarios

These scenarios reflect common needs in each industry and show how each theme starts from one use case. Company names are withheld, and outcomes are rollout targets confirmed through PoC validation.

Electronics manufacturer, mid-sized

NOVARK Audit

Challenge: Preparing for ISO 27001 certification for the first time with a two-person security team, and policies and evidence spread across departments.
Approach: Start from ISMS scope and risk assessment, let AI draft the SoA and four-tier documents, then use the audit assistant for mock Q&A, evidence inventory and an audit pack.
Target: 50% less time to build the system, 60% less time to prepare audit material

Precision components maker, export customers

NOVARK Doc

Challenge: Customer specifications change often, and engineering and QA compare versions by hand before converting them into inspection criteria.
Approach: Gather each specification into one version chain, compare versions clause by clause, and convert customer specs into internal spec sheets and inspection criteria linked to the original clauses.
Target: 50% less time on comparison and on conversion

Chemical plant operations, multiple sites

NOVARK Inspect

Challenge: Equipment rounds are on paper, anomaly photos sit on personal phones and fixes are hard to follow up.
Approach: Move scheduling and rounds to mobile devices with offline sync, let AI review photos and readings first, and route anomalies through to verified closure.
Target: 100% digital inspection records and fully traceable fixes

Service group, several subsidiaries

NOVARK Finance

Challenge: Large volumes of expense receipts are only sampled, and each subsidiary's cash position is visible only at month-end.
Approach: Extract receipts with OCR and check every transaction, combine receivables, payables and bank data for cash-flow alerts, and keep human review at the end.
Target: Full-population review instead of sampling, and earlier warning of cash gaps

Equipment service provider, nationwide

NOVARK Service

Challenge: Phone and LINE requests are handled separately, agents search contracts and manuals for answers, and dispatch details get lost.
Approach: Link phone, LINE and web requests under one case number, answer from contracts and manuals with sources, and create dispatch tickets automatically.
Target: Calls, tickets and engineer reports tracked in one case

Manufacturing group, cross-site management

NOVARK Command

Challenge: Each plant reports figures differently, and preparing for management meetings takes days of reconciling numbers that still disagree.
Approach: Agree one definition and source per metric, connect ERP, MES and CRM to live dashboards, and let managers ask questions in plain language with the calculation shown.
Target: 100% of managed metrics with one definition and source

NAMES / 05

English product names

NOVARK English product names follow one pattern: NOVARK in capitals, followed by one English word with an initial capital, separated by a space and never hyphenated.

ISO management system and audit	NOVARK Audit
Intelligent document management	NOVARK Doc
Intelligent inspection	NOVARK Inspect
Intelligent finance	NOVARK Finance
Intelligent customer service	NOVARK Service
Intelligent war room	NOVARK Command

FAQ / 06

FAQ

What does a NOVARK rollout look like?

Five phases: scenario review, environment setup, PoC validation, go-live and expansion. Start with one theme, validate on real data, then expand to other themes or units.

What hardware does on-premises deployment need?

Three tiers: one GPU server for a PoC or single-department pilot; multiple data-center GPUs when several departments use it; multiple GPU servers with redundancy for company-wide use. Exact configuration follows the scenario review.

Can we try NOVARK without GPU servers?

Yes. A PoC can run on existing equipment, a test environment or a cloud model, and you decide on on-premises hardware after seeing the results.

Can NOVARK support our security audits?

Yes. Its access, traceability and data controls map to common ISO/IEC 27001 and personal data protection requirements, and architecture and control documentation is available for audits and vendor assessments.