KAITIAKI IT
// Secure Portal Architecture · Production Vision v3.0

Kaitiaki
Sentinel

Australia's secure, guardrail-enforced threat intelligence portal.
Built specifically for the Privacy Act 2026.

NVIDIA NIM Powered
NeMo Guardrails
Sovereign Secure Portal
// 01 — The Precipice

The Privacy Act Deadline.
10 December 2026.

Every upload of corporate data to ChatGPT, DeepSeek, or Gemini is a potential APP 1.7 violation. The penalty is $50M. ACSC Essential Eight now mandates AI controls.

APP 1.7 Accountability

The penalty is the greater of $50M, 3x the benefit obtained, or 30% of adjusted turnover. The Australian mid-market (50+ staff) is completely blind to their exposure.

78%

of Australian employees use
unauthorised Generative AI

// 02 — The Market Vulnerability

Why US SaaS Fails Here

Global alternatives (Varonis, Microsoft Purview) are generic, opaque US-based SaaS platforms. Sending local Australian network logs offshore to a US datacenter for compliance checking is an inherent violation of the very data sovereignty principles they claim to protect.

US SaaS = Data Egress Risk
$50–200K/year pricing

The Sovereign Portal Advantage

Kaitiaki Sentinel gives clients the massive computational power of NVIDIA NIM without the extreme hardware overhead of building an on-premise AI deployment.

  • Zero Infra Overhead: Immediate time-to-value.
  • Guardrailed: Inputs completely sanitized.
  • AU-Native: Pure Privacy Act mapping.
// 03 — The Ecosystem

The Full Kaitiaki Suite

Three integrated SaaS products forming a single, secure sovereign portal.

KAITIAKI PORTAL

SSO & Institutional Memory

KAITIAKI SENTINEL

Cloud Threat Detection

KAITIAKI ACADEMY

Targeted Remediation

Sentinel identifies the gap → Academy automatically certifies the user to close it.

// 04 — Core Engineering

Under the Hood: Portal Architecture

Transport & Ingestion

All LLM transport securely routes through LiteLLM. The portal handles CSV, ZIP, XML, Syslog, JSON, and raw PCAP.

/api/audit
/api/report/stream
/api/guardrails/log

Secure Audit Persistence

Every run is written to an encrypted cloud SQLite volume. We preserve the timestamp, user, risk score, JSON findings, and full report text.

INSERT INTO audit_log (timestamp, user, risk_score, findings, raw_report)
// 05 — LLM Architecture

Smart Model Routing

We do not use brute force. Every LLM call in the Sentinel portal is dynamically routed to the highly-specialized model best suited for that exact task.

kaitiaki-nvidia-fast

llama-4-maverick-17b: Sub-second NL-to-pandas, quick threat classification, and critical NeMo guardrail decisions.

kaitiaki-nvidia-coder

qwen3-coder-480b: Optimized for structured data. Executes compliance control ID mapping and JSON output generation.

kaitiaki-nvidia-think

kimi-k2-instruct: Built for structured multi-step reasoning. Deep attack chain analysis and threat actor capability assessment.

kaitiaki-nvidia-main

nemotron-super-49b: The storyteller. Generates elite-quality executive narratives, board reports, and OAIC drafts.

// 06 — Defense Layer

NeMo Guardrails implementation

This makes Kaitiaki the industry leader. A concrete implementation via config.yml and sentinel.co. Absolute transparency for CISO clients.

1. RAW LOG UPLOAD

Messy network logs containing IP addresses, usernames, and potential prompt injections.

2. INPUT RAILS

no_prompt_injection: Strips overrides.
data_minimisation: Hashes PII via sanitize_df.

3. NIM INFERENCE

Task-routed LLM processing (llama-4, qwen3, kimi, nemotron) on sanitized data.

4. OUTPUT RAILS

grounded_findings: Checks log presence.
no_hallucinations: Validates Control IDs.

// 07 — Threat Detection

Advanced Detection Engine

Lateral Movement

Detects internal subnet hop patterns, rapid lateral mapping, and credential stuffing.

Ransomware Precursors

Identifies shadow copy deletion signatures and rapid sequential file operations.

Geo-IP Risk

Cross-references traffic against sanctioned lists (e.g., PRC Intelligence Law flags).

C2 Beacon Detection

Identifies regular-interval outbound communication to unidentified destinations.

Insider Threat Scoring

Combo scoring: After-hours access + Bulk data transfer + Privilege escalation.

Shadow AI Access

Volume metrics on data sent to OpenAI, Anthropic, DeepSeek, and Gemini.

// 08 — Framework Mapping

The map_compliance() Engine

Every finding is automatically tagged with specific control IDs across five major frameworks simultaneously.

  • 01

    AU Privacy Act 2026

    Maps APP 1.7/8/11. Assesses NDB Scheme triggers. Auto-drafts OAIC letters.

  • 02

    ACSC Essential Eight

    Calculates Maturity Levels (1-3) for App Control, Patching, and MFA.

  • 03

    ISO 27001:2022

    Full Annex A control mapping. Identifies SoA gaps from findings.

  • 04

    SOC 2 Type II

    Trust Services Criteria readiness scoring mapped to network behaviors.

  • 05

    PCI-DSS v4.0

    Automatically triggered when cardholder data patterns are detected.

// 09 — User Interface

The Dark SOC Portal

Accessed securely via web browser. Pure performance and clarity without client-side installation.

RESULTS DASHBOARD // APP 1.7 EXPOSURE EXPORT PDF
CRITICAL EXPOSURE
84
$3.2M AUD RISK
FRAMEWORK MAPPING
Privacy Act 2026
85%
Essential Eight
40%
ISO 27001:2022
60%
// 10 — Commercial Model

Pricing Strategy

Starter
Audit Scan
$2.5K
per engagement
  • Single log source
  • APP 1.7 + E8 mapping
  • 90%+ gross margin
Channel
Wholesale
$12K
per year for MSPs
  • One contract = 50 clients
  • MSPs resell at $2-3k/client
  • Massive partner margin
Training
Academy
$499
per seat / year
  • Closes Sentinel gaps
  • 5 active courses
  • Break-even: 157 seats
// 11 — Execution

90-Day Go-To-Market Roadmap

Target Profiles

T1: Compliance SME (Days 1-60). 50-200 staff. Free scan on Palo Alto logs via the portal closes the deal.

T2: MSPs (Days 30-90). Serving 10-50 clients. A single contract covers them all.

T3: Enterprise (Days 60-180). 200+ staff seeking a secure, guardrailed AI portal.

The Timeline

Days 1–30: Direct Pipeline

3 audits @ $2,500. Pitch CISOs on LinkedIn.

Days 31–60: Channel Scale

Convert to 2 annual accounts + 1 MSP wholesale ($54K ARR).

Days 61–90: Expansion

Target: 8 annual + 2 MSP = $195k ARR.

// 12 — Sweat Equity & Asset Value

Built for Production.
Ready today.

Sentinel is a fully realized secure portal architecture, complete with automated compliance mappers, robust SQLite auditing, and production-ready NVIDIA NIM integration.

The Investment Calculation

  • 01
    Development Time: 180 Days (6 Months)
  • 02
    Cadence: 8 hours a day dedicated focus
  • 03
    Rate: $70/hr applied valuation
  • 04
    Compute: $3,000 in AI credits consumed
TOTAL R&D VALUE INVESTED
$103.8K

180 days × 8 hours × $70/hr = $100,800

+ $3,000 AI Infrastructure

KAITIAKI IT
// The Deadline is Near

The Privacy Act takes effect
10 December 2026.

Every week without visibility is a week of unquantified exposure. We have seven months to own this market.

gino.rangitaawa@kaitiaki-it.com
kaitiaki-it.com
KAITIAKI SENTINEL — CONFIDENTIAL