AI Models

AI Models

Overview

NexusEdge Hailo Edition embeds 150+ pre-trained neural models across 5 architectures, compiled for both Hailo-8 and Hailo-10H silicon. All models run on-device with sub-millisecond inference latency. The runtime is chip-aware: it auto-detects which Hailo variant is installed and loads the correct HEF binary.

Model Architectures

Nehebkau (LSTM)

• Purpose: Fault Detection and Diagnostics (FDD)
• Architecture: LSTM recurrent neural network
• Models: 24 equipment-specific variants (H8 + H10 = 48 HEF files)
• Output: Health score (0-100%), 6 fault category scores

Medjed (GRU)

• Purpose: Anomaly detection and predictive maintenance
• Architecture: GRU recurrent neural network
• Models: 24 equipment-specific variants (H8 + H10 = 48 HEF files)
• Output: Anomaly score (0.0-1.0), confidence, prediction horizon

Thoth (HVAC Fault Classifier)

• Purpose: Equipment-specific fault classification
• Architecture: Multi-class classifier
• Models: 24 equipment-specific models
• Categories: 42 fault categories spanning mechanical, electrical, refrigerant, airflow, and sensor failures
• Output: Top-N fault classifications with confidence scores

Anubis (SIEM Threat Detector)

• Purpose: Security Information and Event Management threat detection
• Architecture: Neural threat classifier
• Categories: 34 threat categories including brute force, lateral movement, data exfiltration, protocol anomalies, and insider threats
• Output: Threat category, severity score, recommended action
• Integration: Feeds into NexusShield SIEM export (Syslog, Elasticsearch, Splunk, webhook)

Sage BOW (Chat Retriever)

• Purpose: NexusOracle knowledge retrieval
• Architecture: Bag-of-Words retriever
• Corpus: 617 HVAC diagnostic scenarios
• Output: Top-5 ranked knowledge base entries per query

Equipment Coverage

Each of the 24 equipment types has Nehebkau, Medjed, and Thoth model variants:

AHU, RTU, Boiler, Cascade Boiler, Residential Boiler, Residential Furnace, Chiller, Water-Cooled Chiller, Cooling Tower, Pump, Booster Pump, VFD Pump Pack, Heat Pump, Residential Heat Pump, DOAS, Zone Reheat, Commercial Lighting, Residential Electrical, Garden, Smart Home, Pool Gas Heater, Pool Heat Pump, City Water, Well Water

Chip-Aware Loading

NexusEdge auto-detects the installed Hailo accelerator at startup:

Chip	TOPS	PCIe	HEF Suffix
Hailo-8	26	Gen3 x4	_h8.hef
Hailo-10H	40	Gen4 x4	_h10.hef
None	--	--	CPU fallback

When no NPU is detected, inference falls back to CPU execution. The model selection is fully automatic with no user configuration required.

Inference Pipeline

1. Sensor data collected every 5 seconds via I2C
2. Feature vector built from equipment inputs
3. Nehebkau + Medjed + Thoth inference every 10 seconds
4. Anubis scans system events continuously
5. Results stored in local AegisDB
6. Results pushed to AN Console AegisDB every 60 seconds (Pro tier and above)

On-Controller Training

The AxonML training pipeline allows fine-tuning models with your own operational data. Training runs on-device and the updated weights stay on the controller.

Whitepapers

• Nehebkau FDD Whitepaper
• Medjed Anomaly Detection Whitepaper