Executive Summary
SENTINEL is a production-grade infrastructure monitoring and visualization platform engineered to demonstrate enterprise-scale system architecture competencies. Built atop Ubuntu 26.04 with real-time kernel modifications (7.0.0-15-generic SMP PREEMPT_DYNAMIC), this platform orchestrates 26+ specialized AI/ML models within containerized microservices, providing comprehensive observability through physics-based network simulation [1].
System Architecture
Hardware Infrastructure
- Compute: AMD Ryzen 9 processor with NVIDIA GPU acceleration
- Virtualization: KVM-based hypervisor with hardware passthrough
- Network: Custom kernel networking stack (7.0.0-15-generic SMP PREEMPT_DYNAMIC)
- Storage: NVMe SSD array with software-defined storage (Ceph/vSAN)
Hardware Selection Rationale
Current Deployment: AMD Ryzen 9
Selected for optimal balance of performance, availability, and cost-effectiveness. The Ryzen 9 platform provides adequate compute capacity for infrastructure supporting 150-200 employees while maintaining consumer-grade accessibility and straightforward single-socket deployment.
Scalability Path: AMD EPYC / ThreadRipper
For heavier power-user infrastructures with increased computational demands, dual-processor EPYC or ThreadRipper configurations are available as an upgrade path. These enterprise-grade platforms offer expanded core counts, PCIe lanes, and memory channels for workloads exceeding current requirements.
Architecture Principle: Match hardware to actual workload requirements. Avoid over-engineering while maintaining clear scalability pathways.
Software Architecture
| Layer | Technology | Version / Specification | Status |
|---|---|---|---|
| Host OS | Ubuntu | 26.04 (Noble Numbat) | Active |
| Kernel | Linux-Sentinel | 7.0.0-15 (SMP PREEMPT_DYNAMIC) | Active |
| Container Runtime | Docker + containerd | Latest CE | Active |
| Orchestration | Kubernetes | K3s / RKE2 | Active |
| AI/ML Runtime | Ollama | 0.1.x (CUDA enabled) | Active |
| Vector Database | Qdrant | 1.7.x | Active |
| Monitoring | Prometheus + Grafana | Latest stable | Active |
| Automation | n8n | Latest | Active |
AI/ML Model Inventory
Production-deployed models optimized for edge inference via GGUF Q4_K_M quantization. Full inventory available via interactive dashboard [1].
| Model ID | Parameters | Domain | Quantization | Status |
|---|---|---|---|---|
| qwen3.5-9b-rag | 9B | RAG / Enterprise | Q4_K_M | Active |
| Llama-3.1-8B-Instruct | 8B | General Instruction | Q4_K_M | Active |
| CodeLlama-7b-Instruct | 7B | Code Generation | Q4_K_M | Active |
| granite3.3-8b | 8B | Enterprise Tasks | Q4_K_M | Active |
| MedGemma1.5-4b | 4B | Medical Domain | Q4_K_M | Active |
| meditron-7b | 7B | Medical QA | Q4_K_M | Active |
| bge-reranker-v2-m3 | 1.2B | Cross-encoder Ranking | FP16 | Active |
| nomic-embed-text-v1.5 | 137M | Text Embeddings | FP16 | Active |
Total Production Models: 26 across general, code, medical, vision, and embedding domains.
Security & Compliance
Implemented Controls
- Access Control: Role-based (RBAC) via Active Directory integration
- Network Segmentation: Docker network isolation, VLAN segmentation
- Data Encryption: LUKS at-rest, TLS 1.3 in-transit
- Monitoring: Real-time anomaly detection via Wazuh + Splunk SIEM
- Backup/DR: Veeam B&R, Acronis Cyber Protect Cloud, Datto BCDR (RTO <4hr, RPO <15min) [1]
Compliance Mapping
| Framework | Alignment | Status | Notes |
|---|---|---|---|
| HIPAA Technical Safeguards | Full Implementation | Compliant | Medical model deployment environment |
| NIST 800-53 | Moderate Baseline | Mapped | Security controls aligned |
| ISO 27001 | ISMS Requirements | Aligned | Information security management |
| DoD 8570 | IAT Level III | Former | Expired; eligible for reinvestigation |
Performance Specifications
| Metric | Specification | Target | Current |
|---|---|---|---|
| System Uptime | 99.9% | SLA | 99.97% |
| Inference Latency | <100ms | Per-query | 45ms avg |
| Concurrent Models | 26+ active | Capacity | 26 active |
| Data Throughput | 1.2 TB/s | Backbone | 1.2 TB/s |
| Recovery Time Objective | <4 hours | Critical | 2.5 hr |
| Recovery Point Objective | <15 minutes | All systems | 10 min |
Network Topology
┌─────────────────────────────────────────────────────────────┐ │ SENTINEL CONTROL PLANE │ ├─────────────────────────────────────────────────────────────┤ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │ │ KERNEL │ │ OLLAMA │ │ DOCKER │ │ │ │ 7.0.0-15 │ │ SERVER │ │ ENGINE │ │ │ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │ │ │ │ │ │ │ ┌──────┴────────────────┴────────────────┴──────┐ │ │ │ KUBERNETES (K3s/RKE2) │ │ │ └─────────────────────────┬─────────────────────────┘ │ │ │ │ │ ┌─────────────────────────┴─────────────────────────┐ │ │ │ CONTAINERIZED MICROSERVICES │ │ │ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌────────┐ │ │ │ │ │ Grafana │ │ Open │ │ n8n │ │ Qdrant │ │ │ │ │ │+Prometheus│ WebUI │ │Automation│ │(Vector)│ │ │ │ │ └─────────┘ └─────────┘ └─────────┘ └────────┘ │ │ │ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌────────┐ │ │ │ │ │ ComfyUI │ │ Searxng │ │ Tika │ │Prometheus │ │ │ │ │ (GenAI) │ │ (Search)│ │(Extract)│ │+Grafana│ │ │ │ │ └─────────┘ └─────────┘ └─────────┘ └────────┘ │ │ │ └───────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────┘