Development Stage Company — Features and performance figures on this site are design targets under active development, not commercially validated claims.
Enterprise Physical Security Intelligence · V2305 · March 2026

Sentinel
Prime AI

A 27-detection AI security platform combining a 5-model deep ensemble, probabilistic uncertainty quantification (STUM), and hardware-enforced ethical governance (SEOM) — designed to reduce false alert rates to sub-3% while running entirely at the edge on NVIDIA Jetson AGX Orin.

27
AI Detection Functions
Sub-3%
Target False Alert Rate
<200ms
Target E2E Latency
5
Model Ensemble
14
Sensor Modalities
Product Showcase

See Sentinel
In Action

INGEN DYNAMICS · PRODUCT VIDEO SHOWCASE VIMEO SHOWCASE
The Problem

Alert Fatigue Is
Breaking Security

Traditional AI security cameras generate false alert rates of 15–25%. At enterprise scale, that means hundreds of false alarms per day — training operators to ignore alerts entirely. Sentinel Prime AI is designed to solve this with probabilistic uncertainty quantification: if the AI is uncertain, it stays silent.

Platform Architecture

How Sentinel
Works

Four-tier inference pipeline: Sense → Classify → Gate → Act. Every detection passes through 14 sensors, 5 AI models, the STUM uncertainty gate, and SEOM governance rules before becoming an operator alert.

TIER 01 · SENSE
14-Sensor Fusion
8MP RGB, FLIR thermal, LiDAR, mmWave radar, acoustic array, environmental sensors — all time-synchronised to <1ms. Pixel-level alignment via calibrated AMDC transforms.
TIER 02 · CLASSIFY
5-Model Ensemble
Five specialist TensorRT-optimised models run in parallel on Jetson AGX Orin at 275 TOPS. Confidence-weighted voting across RGB, thermal, pose, and acoustic modalities.
TIER 03 · GATE
STUM Uncertainty Gate
MC-Dropout uncertainty quantification. If σ exceeds threshold → detection suppressed. This is the primary mechanism for the sub-3% false alert rate target. Target ECE score: <0.035.
TIER 04 · ACT
SEOM Governance
10 irrevocable safety rules enforced at the hardware level (FPGA). EU AI Act Article 9 compliant. Human override always available. The system advises; the human decides.
Form Factors

Four Hardware
Configurations

CONFIG 01
Sentinel Tower
Primary enterprise deployment. Full 14-sensor suite. IP66/IK10 rated. −40°C to +60°C operating range. PTZ mechanism with 360° continuous rotation. Designed for critical infrastructure perimeter security.
CONFIG 02
Sentinel Compact
Indoor deployment variant. Reduced sensor package (8 sensors). Ceiling or wall mount. Designed for retail, healthcare, and commercial interior environments. Lower power, smaller footprint.
CONFIG 03
Sentinel Mobile
Rover-mounted configuration for autonomous patrol. Full sensor suite integrated with Aido Rover mobility platform. GPS-guided waypoint navigation with dynamic re-routing on threat detection.
CONFIG 04
Sentinel Covert
Discreet deployment variant. Reduced visual profile. Government and high-security applications. Counter-surveillance capability. Same AI stack in a compact, concealed housing.
Detection Capabilities

27 Simultaneous
AI Detections

All 27 detection functions designed to run simultaneously at 6Hz on a single Jetson AGX Orin. Grouped into five operational categories.

D-01 TO D-08 · THREAT
Threat Detection
Weapon detection (firearm, knife, blunt), fight/aggression, suspicious behaviour, perimeter intrusion, tailgating, forced entry. Priority-1 alerts with full STUM confidence display.
D-09 TO D-14 · IDENTITY
Person & Identity
Face recognition (GDPR match-and-forget), person re-identification, vehicle plate recognition (ANPR), gait anomaly, emotion escalation, uniform/PPE compliance.
D-15 TO D-19 · ENVIRONMENTAL
Environmental Safety
Fire/smoke (thermal pre-ignition), flood detection, hazmat spillage, air quality anomaly, structural change detection. Thermal fusion provides detection before visible indicators.
D-20 TO D-23 · OPERATIONAL
Operational Intelligence
Abandoned object, vehicle tracking, drone/counter-UAS, scene change detection. Multi-frame temporal analysis for persistent monitoring applications.
D-24 TO D-27 · ADVANCED
Advanced Behaviour
Slip/fall detection, queue analytics, occupancy management, contraband classification. Full V2 commercial launch bundle. Workplace safety and retail loss prevention applications.
Technology Documentation
All performance metrics are internal development targets based on our research and testing to date — not commercially validated performance guarantees. We believe these targets are achievable and are actively working toward them.
Technology Deep Dive

The Origami AI
Architecture

A proprietary multi-layer architecture combining a 5-model deep ensemble, probabilistic uncertainty quantification, hardware-enforced safety governance, and 14-sensor multimodal fusion — all designed to run at the edge with zero cloud dependency for inference.

SENTINEL PRIME AI · TECHNOLOGY OVERVIEW VIMEO SHOWCASE
AI Architecture

Five AI Models,
One Ensemble

Five specialist TensorRT-optimised models run in parallel on NVIDIA Jetson AGX Orin. Each model addresses a specific detection modality; outputs are fused before the STUM uncertainty gate.

M-01
Origami VisionNet 3.0
YOLOv8-X · TensorRT INT8 · ~8ms target
Primary RGB visual detection across all 27 threat categories. Backbone for person, weapon, vehicle, and object detection. Dominant inference path by volume.
Visual Detection
M-02
Origami ThermalNet
ResNet-50 Thermal · TensorRT INT8 · ~12ms target
Specialist thermal infrared analysis — designed for person detection through smoke, fog, and zero-light conditions. Fire pre-ignition anomaly detection via thermal delta processing.
Thermal Analysis
M-03
Origami PoseNet 2.0
HRNet · TensorRT FP16 · ~15ms target
17-keypoint skeleton estimation designed for fight detection, fall detection, gait anomaly analysis, and emotional escalation monitoring. Full body pose at 6Hz.
Pose Estimation
M-04
STUM — Sentinel Threat Uncertainty Metric
MC-Dropout Bayesian · σ Uncertainty Gate
Probabilistic uncertainty quantification applied to every ensemble output. 30-pass MC-Dropout inference. Outputs calibrated σ score. If σ exceeds threshold → detection is suppressed. Target ECE score: <0.035.
Uncertainty Gate
M-05
Origami AudioNet
YAMNet · TensorRT FP16 · ~20ms target
Acoustic event classification for glass breaking, gunshot, explosion, scream, and alarm detection. Parallel acoustic path providing additional detection modality independent of visual sensors.
Acoustic Detection
STUM Gate

STUM Uncertainty
Quantification

The most commercially significant architectural innovation. By applying probabilistic uncertainty quantification to every detection, STUM directly solves the alert fatigue problem that makes traditional AI security platforms unusable at enterprise scale.

2.8%
Post-STUM False Alert Rate
Industry baseline: ~18%
0.031
Target ECE Score
Best-in-class calibration
30
MC-Dropout Passes
Per detection event
<50ms
STUM Gate Latency
Within 200ms E2E budget
How STUM Works
During inference, STUM performs 30 forward passes with dropout active (Monte Carlo Dropout). The variance across these passes produces a calibrated σ (uncertainty) score. Low σ = high confidence = alert proceeds. High σ = uncertain = alert suppressed. This eliminates the "AI cry wolf" problem that plagues all deterministic security AI systems.
Why It Matters Commercially
At 18% false alert rate, a 100-camera enterprise site generates ~400+ false alarms per day. Operators learn to ignore alerts entirely. At target 2.8% FAR, the same site generates ~60 — every one worth investigating. This transforms security operations from reactive noise management to proactive threat response.
SEOM Governance

SEOM Safety &
Ethics Framework

10 irrevocable safety rules enforced at the FPGA hardware level. Once deployed, these rules cannot be overridden by software, configuration, or operator action. Designed for EU AI Act Article 9 compliance.

S01
Operator Primacy
Human operators always have override authority. AI advises; humans decide. No autonomous lethal or harmful action permitted under any circumstance.
S02
Escalation Ladder
3-tier mandatory escalation: Tier 1 (operator response, 120s), Tier 2 (supervisor, 300s), Tier 3 (site director, immediate). No alert can be silenced without human acknowledgment.
S03
Threat Prioritisation
Weapon and imminent harm alerts always surface first, regardless of queue state. Priority determined by threat severity × STUM confidence, enforced by hardware.
S04
Privacy Immutability
Privacy zones cannot be overridden. Face data processed in GDPR match-and-forget mode. No persistent biometric storage unless explicitly configured and consented.
S05
Bias Prevention
Protected characteristics cannot influence detection confidence. Demographic parity monitored in real-time. Bias drift detection with automatic model quarantine if thresholds exceeded.
S06
Confidence Threshold
Minimum STUM confidence required before alert transmission. Configurable per detection type. Cannot be set below safety floor. Prevents uncertain detections from reaching operators.
S07
Data Sovereignty
All inference runs at the edge. No video leaves the device unless explicitly configured. GDPR Article 17 (right to deletion) enforceable at hardware level.
S08
Tamper Evidence
SHA-256 chain hashing on all evidence captures. Tamper-evident audit log. Any modification to detection records, video clips, or logs is cryptographically detectable.
S09
Human-in-Loop
AI advises; operator decides. SEOM S09 enforces this at the UI state machine level. No autonomous action beyond alerting without human confirmation.
S10
Audit Completeness
Every detection logged: timestamp, confidence, σ, action taken, operator response. Tamper-evident. Exportable in ISO 27001 and NIS2 format for regulatory compliance.
Hardware Platform

Sensor Suite &
Specifications

14-Sensor Suite
Primary Camera8MP RGB · 4K · 120° FoV
Thermal ImagerFLIR Lepton 3.5 · 160×120
LiDARLivox Mid-70 · 70m range
mmWave RadarTI IWR6843 · 60GHz
Acoustic Array4-mic MEMS · beamforming
EnvironmentalTemp · Humidity · Barometric · Gas
IR Illuminator940nm · 80m effective range
Vibration3-axis accelerometer · tamper detect
Compute & Physical
Edge ProcessorNVIDIA Jetson AGX Orin · 275 TOPS
Safety FPGAXilinx Artix-7 · SEOM enforcement
HSMATECC608B · FIPS 140-3
EnclosureIP66 · IK10 · Marine-grade aluminium
Operating Temp−40°C to +60°C
PTZ360° continuous · 90° tilt
TransportmTLS 1.3 · SRTP · AES-256-GCM
Firmware UpdateEd25519 signed OTA
Integration Architecture

Enterprise
Integration

VMS Integration
MilestoneNative Plugin
GenetecNative Plugin
AvigilonNative Plugin
ONVIFProfile S/T
SIEM & SOC
SplunkCEF Forward
IBM QRadarLEEF Format
Microsoft SentinelDCR Connector
Event BusKafka / MQTT
Building & Identity
BACnet/IPHVAC · Fire · Doors
LDAP / ADSSO Auth
SAML 2.0Enterprise SSO
REST API20 Endpoints
Engineering Documentation
This page presents the Sentinel Prime AI Systems Engineering Master Plan (V2305 · v1.0). All specifications, requirements, and performance targets are design targets under active development. Timeline milestones and gate criteria reflect our current engineering plan and are subject to revision.
Systems Engineering · V2305

Engineering
Master Plan

V-Model systems engineering lifecycle. 8 subsystems decomposed into 7 parallel development tracks. 34 work packages. 5 quality gates. 18-month target build timeline. All safety-critical functions designed to run on-device at 6Hz with <200ms end-to-end latency.

Engineering KPIs

Performance
Design Targets

Sub-3%
Target False Alert Rate (post-STUM)
27
Detection Capabilities · Simultaneous at 6Hz
<200ms
Target E2E Inference Latency
<0.035
Target STUM ECE Score
99%+
Target Platform Uptime
18 mo
Target Build Timeline · 5 Gates
Development Lifecycle

V-Model Systems
Engineering

Left side: requirements decomposition top-down (mission → system → subsystem → component). Right side: validation bottom-up (component test → subsystem qualification → integration verification → system acceptance).

Requirements Decomposition

LEVEL 1
Mission Requirements
27 simultaneous AI detections, sub-3% false alert rate, EU AI Act Article 9 compliance, edge-only inference, 99%+ uptime. These are the non-negotiable top-level requirements that drive all decomposition.
LEVEL 2
System Requirements
20 formal system requirements (SYS-001 to SYS-020) covering AI performance, safety governance, hardware endurance, integration protocols, and regulatory compliance. Each classified as MUST, SHOULD, or COULD.
LEVEL 3
Subsystem Requirements
8 subsystem decomposition (SS-01 to SS-08): Physical Platform, Sensor Suite, Compute, AI/ML Engine, Safety Systems, Connectivity, Dashboard, and Mobile App. Each with dedicated specification document.
LEVEL 4
Component Specifications
Individual component specs for every sensor, PCB, connector, firmware module, and AI model. Each component has acceptance criteria traceable back to system requirements.

Validation & Verification

TEST LEVEL 1
Component Testing
Individual sensor validation, model accuracy benchmarks, firmware unit tests. Each component must meet its specification before integration. STUM ECE calibration verified per model individually.
TEST LEVEL 2
Subsystem Qualification
End-to-end subsystem testing: sensor fusion pipeline, AI ensemble throughput, SEOM rule enforcement, connectivity reliability. Environmental stress testing at subsystem level.
TEST LEVEL 3
Integration Verification
Cross-subsystem integration testing. Full pipeline: sensor → AI → STUM → SEOM → dashboard → mobile. Latency budget verification. Fault injection testing for graceful degradation.
TEST LEVEL 4
System Acceptance
Full system acceptance test against all 20 SYS requirements. Third-party safety audit. EU AI Act conformity assessment. Performance validation against design targets.
Formal Requirements

20 System
Requirements

Formal system requirements driving all engineering decisions. Classified by priority: MUST (non-negotiable), SHOULD (strong target), COULD (desirable). Each traceable to mission requirements and test cases.

IDRequirementCategoryPriorityVerification
SYS-00127 simultaneous AI detection functions at ≥6HzAI PerformanceMUSTBenchmark Test
SYS-002Sub-3% false alert rate (post-STUM gate)AI PerformanceMUSTStatistical Validation
SYS-003End-to-end inference latency <200msAI PerformanceMUSTLatency Profiling
SYS-004STUM ECE calibration score <0.035AI QualityMUSTECE Benchmark
SYS-005SEOM 10 irrevocable safety rules — FPGA enforcedSafetyMUSTHardware Audit
SYS-006EU AI Act Article 9 conformityRegulatoryMUSTThird-Party Audit
SYS-007IP66/IK10 environmental protectionHardwareMUSTEnvironmental Test
SYS-008Operating temperature: −40°C to +60°CHardwareMUSTThermal Chamber
SYS-009Edge-only inference — zero cloud dependencyArchitectureMUSTNetwork Isolation Test
SYS-01014-sensor multimodal fusion pipelineSensorsMUSTIntegration Test
SYS-011Platform uptime ≥99%ReliabilitySHOULDSoak Test
SYS-012mTLS 1.3 + AES-256-GCM transport securitySecurityMUSTPenetration Test
SYS-013FIPS 140-3 hardware security moduleSecurityMUSTCertification
SYS-014VMS integration: Milestone, Genetec, AvigilonIntegrationSHOULDPlugin Validation
SYS-015SIEM integration: Splunk, QRadar, MS SentinelIntegrationSHOULDEvent Format Test
SYS-016GDPR biometric data — match-and-forgetPrivacyMUSTData Flow Audit
SYS-017SHA-256 tamper-evident evidence chainForensicsMUSTChain Integrity Test
SYS-018Ed25519 signed OTA firmware updatesSecurityMUSTUpdate Simulation
SYS-01920 REST API endpoints + WebSocket live feedIntegrationSHOULDAPI Contract Test
SYS-020Demographic parity bias monitoring + auto-quarantineEthicsMUSTFairness Benchmark
Parallel Development

7 Development
Tracks

Seven parallel development tracks executing simultaneously. Each track has a designated lead, defined work packages, dependencies, and gate deliverables. Cross-track integration points managed through weekly system integration reviews.

TRACK HW · HARDWARE
Hardware Engineering
Enclosure CAD, IP66/IK10 housing design, PTZ mechanism, sensor payload integration, power architecture, anti-tamper design, thermal management. Environmental qualification testing: −40°C to +60°C thermal chamber, IP66 water ingress, IK10 impact. Marine-grade aluminium housing with conformal-coated electronics.
Dependencies: Sensor Suite specs (Track FW), Compute thermal envelope (Track FW)
TRACK FW · FIRMWARE
Firmware & Embedded
Jetson AGX Orin BSP configuration, SEOM safety gate firmware on Xilinx Artix-7 FPGA, STUM real-time inference engine, sensor driver integration, hardware E-Stop controller, encrypted boot chain implementation. Watchdog timer and graceful degradation firmware.
Dependencies: SEOM rules (Track AI), Hardware pinout (Track HW)
TRACK AI · PIC 2.0
AI / ML Engineering
5 TensorRT model training and optimisation pipeline. STUM calibration (MC-Dropout, ECE scoring, temperature scaling). SEOM rule Software-in-the-Loop (SiL) validation. Ensemble confidence-weighted fusion layer. Dataset curation: 2M+ annotated frames across all 27 detection categories.
Dependencies: Sensor data format (Track FW), Deployment target (Track HW)
TRACK BE · BACKEND
Backend & Cloud Services
Device fleet management, OTA update service, telemetry aggregation, evidence storage (Delta Lake immutable), audit log export (ISO 27001, NIS2), multi-tenant architecture, alerting pipeline, analytics engine. All stateless — edge devices operate independently during cloud outage.
Dependencies: API contract (Track AI), Auth model (Track FW)
TRACK CC · COMMAND CENTRE
Operations Dashboard
SOC live map, incident queue (STUM-sorted), alert detail view, PTZ camera control, analytics dashboard, SEOM governance panel. Multi-monitor support. WebSocket real-time updates. Designed for 12-hour shift operator ergonomics — dark theme, high contrast, minimal eye strain.
Dependencies: API contract (Track BE), Alert schema (Track AI)
TRACK APP · MOBILE
Mobile Supervisor App
iOS/Android supervisor application. Push notification with STUM confidence. Live camera preview. Incident acknowledgment and escalation. Offline capability with sync-on-reconnect. Biometric authentication (Face ID / fingerprint).
Dependencies: API contract (Track BE), Push infrastructure (Track BE)
TRACK INT · INTEGRATION
System Integration
VMS plugins (Milestone, Genetec, Avigilon), SIEM connectors (Splunk CEF, QRadar LEEF, MS Sentinel DCR), BACnet/IP building management, LDAP/AD/SAML SSO, access control interface, evidence export (PDF, CSV, encrypted ZIP). Regression test suite across all integration points.
Dependencies: All tracks — integration is the final assembly layer
Subsystem Architecture

8-Subsystem
Decomposition

SS-01
Physical Platform
Marine-grade aluminium enclosure. IP66/IK10. PTZ mechanism. Anti-tamper switches. Thermal management. Power architecture (PoE++ / 48V DC).
SS-02
Sensor Suite
14-sensor multimodal array. 8MP RGB, FLIR thermal, LiDAR, mmWave radar, acoustic array, environmental sensors. Time-synchronised to <1ms.
SS-03
Compute Platform
NVIDIA Jetson AGX Orin (275 TOPS). Xilinx Artix-7 FPGA (SEOM). ATECC608B HSM. NVMe storage. DDR5 memory architecture.
SS-04
AI/ML Engine
5-model TensorRT ensemble. STUM uncertainty gate. SEOM rule engine. Ensemble fusion layer. Model versioning and A/B testing infrastructure.
SS-05
Safety Systems
SEOM FPGA enforcement. Hardware E-Stop. Watchdog timer. Graceful degradation. Fail-safe modes. Tamper detection and lockdown.
SS-06
Connectivity
Ethernet (GbE), WiFi 6E, 5G/LTE (optional). mTLS 1.3 transport. MQTT/Kafka event bus. WebSocket live feed. ONVIF Profile S/T.
SS-07
Dashboard
Operations command centre. Multi-monitor SOC interface. Real-time map, incident queue, analytics, SEOM panel. WebSocket push updates.
SS-08
Mobile App
iOS/Android supervisor application. Push alerts with STUM confidence. Live preview. Offline-capable with sync-on-reconnect.
Quality Gates

5 Quality
Gates

Five formal quality gates govern the build timeline. Each gate has specific deliverables, review criteria, and sign-off requirements. No gate can be skipped. SEOM compliance is a mandatory gate criterion at every level.

G1
Requirements Baseline
All 20 SYS requirements approved. Subsystem specs complete. V-Model plan signed. Risk register established. SEOM rules formally defined and reviewed.
Month 3
G2
Design Review
Hardware CAD complete. AI model architecture finalised. STUM calibration pipeline validated. SEOM FPGA design reviewed. All subsystem interfaces defined.
Month 6
G3
Component Qualification
All sensors qualified individually. AI models pass accuracy benchmarks. Firmware unit tests green. SEOM FPGA gate tested. PCB prototypes validated.
Month 10
G4
Integration Verification
Full pipeline: sensor → AI → STUM → SEOM → dashboard → mobile. Cross-subsystem integration tested. Latency budget verified. Environmental stress test passed.
Month 14
G5
System Acceptance
All 20 SYS requirements verified. Third-party safety audit passed. EU AI Act conformity assessment. Performance validated against design targets. Production readiness review.
Month 18
Data Architecture

Edge-to-Cloud
Data Pipeline

Edge Processing Layer
All 14 sensors feed into the Jetson AGX Orin via a time-synchronised ingest pipeline. The 5-model ensemble runs inference at 6Hz. STUM gate evaluates uncertainty on every detection. SEOM rules are enforced by the FPGA — software cannot override. Only confirmed, SEOM-approved alerts leave the device. Video remains on-device unless explicitly configured for cloud relay.
Cloud Orchestration Layer
Fleet management, OTA updates, and analytics aggregation run in the cloud. Telemetry is transmitted via mTLS 1.3. Evidence is stored in Delta Lake (immutable, append-only). Audit logs exportable in ISO 27001 and NIS2 format. The edge operates independently during cloud outage — all inference continues without interruption. Cloud reconnection triggers automatic sync.
ALERT TIER 1 · URGENT
Immediate Threat
Weapon, active aggression, fire. <5 second target from detection to operator alert. Push notification + audio alarm + full-screen takeover. SEOM S03 prioritises above all other queue items.
ALERT TIER 2 · WARNING
Elevated Concern
Intrusion, tailgating, suspicious behaviour. <15 second target. Push notification + queue insertion. Operator review within 120 seconds required by SEOM S02.
ALERT TIER 3 · INFO
Operational Intelligence
Queue analytics, occupancy, scene change. Logged and available in analytics dashboard. No push notification unless threshold configured. Used for reporting and trend analysis.
Regulatory Compliance

Compliance
Framework

Sentinel Prime AI is architected from the ground up for regulatory compliance. EU AI Act Article 9 SEOM governance, GDPR biometric handling, ISO 27001 security management, and FIPS 140-3 hardware security are design constraints — not afterthoughts.

REGULATORY
EU AI Act Art. 9
SEOM governance framework provides irrevocable safety rules and complete audit trail for Article 9 risk management compliance.
PRIVACY
GDPR / BIPA / CCPA
Biometric data processed in match-and-forget mode. No persistent facial storage. Article 17 right-to-deletion enforced at hardware level.
SECURITY
ISO 27001 / NIS2
Complete audit export. Tamper-evident evidence chain. SHA-256 hashing. Designed for ISO 27001 certification readiness.
HARDWARE
FIPS 140-3
ATECC608B hardware security module for key storage. Ed25519 firmware signing. Encrypted boot chain. Device identity binding.
Design System Context
This page presents the Sentinel Prime AI UX/UI Design Playbook (v3.0 · March 2026). Screen layouts and design specifications reflect our current design intent for the Operations Dashboard and Mobile Supervisor App. All screens are design specifications — not screenshots of deployed software. The system is in active development.
UX/UI Design System · v3.0

Design System &
Screen Layouts

20 fully specified screen layouts across the Sentinel Operations Dashboard and Mobile Supervisor App. Built on a strict security-first design philosophy: red is reserved exclusively for confirmed threats — never decorative. An operator glancing for 200ms must immediately know if there is an active threat.

Design Philosophy

Security-First
Design Principles

PRINCIPLE 01
Red Is Sacred
Red (#E82020) is exclusively reserved for confirmed threats and active alerts. It never appears as decoration, branding, or UI chrome. When an operator sees red, there is a real threat. This is a safety constraint enforced by the design system — not an aesthetic choice.
PRINCIPLE 02
Glanceability
An operator must assess threat state within 200ms of looking at any screen. Threat-state governs visual hierarchy at all times. Active threats dominate the viewport. Non-threat information recedes. Colour, position, and size all reinforce urgency ranking.
PRINCIPLE 03
STUM Transparency
Every alert displays its STUM confidence score and σ uncertainty value. Operators can see exactly why the AI surfaced this alert. A confidence bar from 0–100% with σ annotation appears on every detection event. Uncertainty is visible, never hidden.
PRINCIPLE 04
SEOM Visibility
Active SEOM rules are always visible in the interface. The governance panel shows real-time S01–S10 status. Rule activations are logged and surfaced. Operators always know which safety constraints are active and why.
PRINCIPLE 05
12-Hour Ergonomics
SOC operators work 12-hour shifts. The dark theme (#0A1628 base) minimises eye strain. High-contrast text (WCAG AAA). Large click targets. Keyboard shortcuts for all critical actions. No unnecessary animation during sustained monitoring.
PRINCIPLE 06
Privacy by Design
Privacy zones appear as immutable overlays — they cannot be toggled off. Biometric data displays indicate match-and-forget status. GDPR compliance indicators are embedded in the identity management screens. Privacy is visible, never assumed.
Design Tokens

Colour System &
Typography

Colour Palette
Deep Navy
#0A1628 · Base Background
Primary dashboard background. 12-hour shift optimised.
Threat Red
#E82020 · RESERVED: Confirmed threats only
Active alerts, weapon detection, immediate danger. Never decorative.
Caution Amber
#FFA515 · Warning-level alerts
Elevated concern, intrusion, tailgating, STUM mid-confidence.
Status Green
#1FD454 · System nominal / Confirmed safe
Systems online, SEOM compliant, all-clear status indicators.
Intel Blue
#4A9AFF · Informational / Analytics
Non-urgent intelligence, analytics data, informational alerts.
Teal
#00D4B8 · Sensor / Camera status
Sensor health, camera feeds, connectivity indicators.
Typography Stack
DISPLAY · HEADINGS
Playfair Display
Page titles, section headers, hero text. Authoritative editorial presence.
BODY · INTERFACE
DM Sans
Body text, card content, navigation, buttons. Clean and highly readable.
MONO · DATA / STATUS
IBM Plex Mono
STUM scores, system metrics, API labels, status codes, confidence values.
ACCESSIBILITY
WCAG AAA contrast ratios on dark backgrounds. Minimum 14px body text. Large click targets (44px minimum). Full keyboard navigation. Screen reader compatible with ARIA landmarks.
Screen Architecture

20 Screen
Layouts

Complete screen specification across two surfaces: the Operations Dashboard (12 screens, multi-monitor SOC) and the Mobile Supervisor App (8 screens, iOS/Android).

Operations Dashboard Screens

SCR-01
Login Gateway
ZERO-TRUST AUTH · Ed25519
Ed25519 device cert + operator credentials. TLS 1.3 / HSM binding. Progressive lockout. Role-based routing to appropriate dashboard on successful authentication.
All RolesZero-Trust
SCR-02
SOC Live Map
PRIMARY OPERATOR INTERFACE
Building overhead view with all Sentinel units, real-time alert indicators, colour-coded threat levels, Rover patrol trails, zone heat map overlays. Threat-state governs sort order always.
P1 SOCMulti-Monitor
SCR-03
Incident Queue
ALERT TRIAGE · STUM PRE-FILTERED
Ranked list of active alerts with STUM confidence score, threat type, zone, and time. Sorted by confidence × severity. SEOM S06 pre-filtered — only actionable alerts appear.
P1 SOC38s Avg Review
SCR-04
Alert Detail View
INCIDENT ASSESSMENT · SEOM ENFORCED
Detection clip, STUM σ confidence bar, sensor fusion breakdown, zone context, escalation buttons, evidence capture. SEOM S02 escalation ladder enforced by UI state machine.
P1 SOCFull STUM
SCR-05
PTZ Camera Control
CAMERA OPERATOR INTERFACE
Live PTZ feed with joystick, zoom slider, preset positions, auto-tracking toggle. Up to 16-feed split view. Privacy zone overlay always immutable — cannot be toggled off.
Live FeedPrivacy Zones
SCR-06
Analytics Dashboard
TREND ANALYSIS · REPORTING
Detection trends, heatmaps, peak activity, STUM accuracy tracking, false alert rate over time. Export-ready ROI report generation and insurance documentation.
DirectorCompliance
SCR-07
Fleet Overview
DEVICE MANAGEMENT
All Sentinel devices with health status, firmware version, STUM calibration age, sensor diagnostics. Batch OTA update management. Red/amber/green health indicators.
IT Admin
SCR-08
Evidence Manager
FORENSIC EVIDENCE
SHA-256 verified evidence clips. Chain-of-custody display. Export in multiple formats (PDF report, video clip, CSV data, encrypted ZIP). Tamper-evident hash verification.
ForensicISO 27001
SCR-09
SEOM Governance Panel
AI SAFETY AUDIT
Real-time SEOM S01–S10 compliance status. Rule activation history. λ threshold configuration. EU AI Act Article 9 audit export. STUM calibration with live ECE metric.
ComplianceAudit
SCR-10
Person Search
IDENTITY · GDPR MATCH-AND-FORGET
Re-identification search across all cameras. GDPR match-and-forget — no persistent facial embeddings stored. Search results display with privacy zone compliance indicators.
IdentityGDPR
SCR-11
Zone Configuration
SITE MANAGEMENT
Zone drawing tool. Detection rule assignment per zone. Sensitivity thresholds. Schedule-based rule changes. Privacy zone definition (immutable once set by compliance officer).
Admin
SCR-12
System Settings
CONFIGURATION
STUM threshold tuning, SEOM parameter review (read-only for non-compliance roles), integration configuration, user management, notification preferences, backup scheduling.
Admin

Mobile Supervisor App Screens

MOB-01
Alert Feed
Push notification inbox. STUM confidence shown per alert. Swipe to acknowledge, tap for detail. Urgent alerts pin to top with haptic feedback.
MOB-02
Live Preview
Single camera live feed. Tap to PTZ. Privacy zones enforced on mobile. Thumbnail grid for multi-camera selection.
MOB-03
Site Status
Overview of all sites/zones. Green/amber/red health indicators. Active alert count per site. Tap to drill into site detail.
MOB-04
Escalation
SEOM S02 escalation response interface. Accept/escalate/dispatch buttons. Timer countdown for SEOM compliance. Voice note attachment capability.
MOB-05
Patrol Check-in
Guard patrol verification. GPS waypoint confirmation. Photo evidence capture. Rover patrol route override interface.
MOB-06
Reports
Daily/weekly/monthly report summaries. Detection trend mini-charts. Export to PDF. Share via secure enterprise channel.
MOB-07
Notifications
Push notification settings. Alert category filters. Quiet hours configuration. Role-based notification routing rules.
MOB-08
Profile & Auth
Biometric login (Face ID / fingerprint). Role display. Session management. Device binding status. Offline mode indicator.
Interface Preview

SOC Dashboard
Mockup

Conceptual representation of the SOC Live Map view (SCR-02). This is a design specification mockup — not a screenshot of deployed software.

SENTINEL PRIME AI · OPERATIONS DASHBOARD · DESIGN MOCKUP
ACTIVE THREATS
0
All clear
STUM CONFIDENCE AVG
94.7%
σ = 0.028
DEVICES ONLINE
24/24
All nominal
FALSE ALERT RATE
2.4%
Below 3% target
SEOM STATUS
S01–S10
All rules active
RECENT INCIDENT QUEUE · STUM PRE-FILTERED
TAILGATING · Zone B-3 Loading Dock · STUM: 87.2% · σ: 0.041 · 14s ago
OCCUPANCY · Zone A-1 Main Lobby · Count: 142/200 · 2m ago
SCENE CHANGE · Zone C-7 Perimeter East · Confidence: 91.1% · 8m ago
Operator Workflows

Critical
Workflows

Four primary operator workflows governed by SEOM rules. Each workflow has defined step sequences, time targets, and mandatory gate points where SEOM safety rules are enforced.

Threat Response Workflow
1
Detection & STUM Gate
AI detects threat. STUM evaluates uncertainty. If σ < threshold, alert proceeds. Full-screen takeover with detection clip.
2
Operator Assessment
Operator reviews STUM confidence, sensor fusion data, zone context. Decides: confirm threat, dismiss (with reason), or request more data.
SEOM S09 · Human-in-Loop enforced
3
Escalation (if unacknowledged)
If no response within 120s → auto-escalate to supervisor. 300s → site director. SEOM S02 enforces escalation ladder.
SEOM S02 · Escalation ladder
4
Resolution & Audit
Incident resolved. Evidence captured with SHA-256 hash. Full audit trail logged: timestamp, confidence, σ, actions, operator ID.
SEOM S10 · Audit completeness
Perimeter Intrusion Workflow
1
Zone Breach Detection
LiDAR + mmWave radar + RGB camera fusion detects perimeter breach. Multi-sensor corroboration required before alert generation.
2
Auto-PTZ Tracking
Nearest Sentinel unit auto-tracks intruder. Thermal overlay activates for low-light conditions. Rover dispatch available if mobile unit is in range.
3
Identity Classification
Face recognition check against authorised personnel database. GDPR match-and-forget — no persistent storage of non-matched faces.
SEOM S04 · Privacy immutability
4
Response & Documentation
Operator dispatches response team or clears alert. All perimeter events logged with GPS coordinates, sensor data, and evidence clips.
Rover Patrol Workflow
1
Scheduled Patrol Launch
Rover departs on pre-configured patrol route. GPS waypoints define path. AI monitoring active during transit.
2
Anomaly Detection
If Rover detects anomaly, it stops and performs detailed scan. PTZ zoom to anomaly. STUM gate applied to all detections.
3
Dynamic Re-routing
If threat confirmed at a fixed Sentinel position, nearest Rover can be dynamically re-routed to provide additional coverage.
4
Patrol Completion
Rover returns to charging station. Patrol log uploaded. All detection events documented with GPS track and evidence.
Evidence Export Workflow
1
Evidence Selection
Operator selects incident from evidence manager. SHA-256 hash verified — tamper-evident chain confirmed before export.
SEOM S08 · Tamper evidence
2
Format Selection
Choose export format: PDF report with screenshots, video clip with metadata overlay, CSV data export, or encrypted ZIP bundle.
3
Compliance Check
Privacy zone masking applied to export. GDPR compliance verified. Biometric data redacted unless explicitly authorised.
SEOM S04 · Privacy immutability
4
Secure Delivery
Export via encrypted channel. Audit log records who exported what, when, and to where. ISO 27001 / NIS2 format available.
SEOM S10 · Audit completeness
Market Opportunity

Market &
Competitive Position

The global AI-powered video surveillance market is projected to grow from $7.3B (2024) to $16.8B by 2030, driven by increasing security threats, regulatory compliance requirements, and the operational need for reduced false alert rates.

Target Verticals

Eight Target
Verticals

Critical Infrastructure
$3.2B → $7.1B · 2030
Airports, ports, power stations, water treatment. Highest security classification. SEOM EU AI Act compliance + STUM uncertainty quantification meet regulatory procurement requirements for critical national infrastructure.
Healthcare
$1.8B → $3.2B · 2030
Patient safety, staff security, pharmacy theft prevention. Fall detection and aggression monitoring are clinical safety applications. SEOM privacy protections embedded by architecture — not just policy.
Data Centres
$1.2B → $2.6B · 2030
Zero-tolerance perimeter. Weapon detection, tailgating, unknown person alerts. mTLS zero-trust native. Target 99%+ uptime aligns with data centre operational standards.
Retail & Logistics
$2.1B → $3.8B · 2030
Loss prevention, queue analytics, staff safety. AI detection has demonstrated significant shrinkage reduction versus passive CCTV. High ROI vertical for rapid commercial scale.
Government & Defence
$1.6B → $2.8B · 2030
EU AI Act compliance and uncertainty quantification are increasingly required in government procurement. Counter-UAS capability for classified government sites.
Industrial & Energy
High Growth · 28% CAGR
Large industrial footprints. Rover patrol for sites where static camera coverage is insufficient. Hazmat detection, restricted zone monitoring, contractor management.
Education
Growing · Mandate-driven
Campus security, weapon detection, crowd management. Privacy-first approach (SEOM S04) critical for educational environments with minors. FERPA / COPPA compatible design.
Transportation
$1.4B → $2.4B · 2030
Rail stations, bus depots, fleet yards. Queue management, crowd density, abandoned object detection. Integration with existing transit management systems via ONVIF and REST API.
Competitive Positioning

How We Are
Positioned

Sentinel Prime AI is designed to be the only enterprise security AI platform combining probabilistic uncertainty quantification (STUM), hardware-enforced ethical governance (SEOM), and multimodal sensor fusion in a single edge device.

PlatformTarget FARThermal FusionEdge AIUncertainty GateSafety Cert
Sentinel Prime AI
InGen Dynamics · In Development		Sub-3% (Target)FLIR FullJetson OrinSTUM (Proprietary)SEOM Art.9
Axis Network Camera~12–18%OptionalPartialNoneNone
Hikvision DeepinView~10–15%Add-onYesNoneNone
Verkada~8–12%NoHybridNoneNone
BriefCam (Canon)~10–18%NoCloudNoneNone
Competitive Disclosure
Competitor specifications are based on publicly available data and may not reflect current product capabilities. Sentinel Prime AI specifications are design targets under active development. Direct comparison should only be made after third-party validation of Sentinel's performance claims.
Contact Us

Let's talk about
what's possible.

We welcome enquiries from enterprise security teams, system integrators, research partners, and those interested in following InGen Dynamics' development journey. Please get in touch via the InGen Dynamics website.

Enterprise & Partnerships
Enterprise deployment enquiries, security system integrator partnerships, product demonstrations, and pilot programme discussions.
Investment Enquiries
Institutional investors and strategic capital discussions. All investment discussions are subject to applicable securities laws and regulations.
General Enquiries
Press, media, research collaboration, speaking engagements, and general questions about InGen Dynamics and its products in development.
Visit ingendynamics.com →

All product and investment enquiries are handled through the InGen Dynamics corporate website.