📦 Sintrones Edge AI Starter Kit

“Edge AI Vision + Sensor Gateway” for Vehicle / Factory / City Use

The Sintrones Edge AI Starter Kit is a production-ready, open-source framework designed to accelerate real-world deployments of AI-enhanced sensor fusion across vehicles, factories, and smart cities.

Built on rugged industrial-grade hardware, it enables seamless integration of vision AI, sensor telemetry, edge dashboards, and protocol adapters (MQTT, Modbus, CANbus) to create deployable proof-of-concepts and real-time systems.

Ideal for system integrators, smart factory teams, and urban solution architects, this repo provides all core modules and examples to quickly demonstrate AI value at the edge.

💡 Sales + Collaboration: Use this as a customer-facing PoC and R&D starter kit. Ideal for OEMs, system integrators, and smart infrastructure pilots in Thailand or SEA deployments.

---

🚀 Features

• 🎥 Multi-Modal Sensor Input — Real camera streams + industrial signals (USB, PoE, RS232, GPIO)
• 🧠 AI Model Inference — Supports YOLOv5, OpenVINO, or ONNX for object detection and event logic
• 📊 Dashboards — Visualize detections and sensor states via Streamlit (lightweight) or Grafana (pro)
• 🔌 Industrial Protocol Support — Communicates via MQTT, Modbus RTU/TCP, and CANBus for machine/vehicle data
• 📡 Mobility-Ready — Integrates 5G modules, GNSS/GPS, and CAN for use in transportation/fleet systems
• 🔄 OTA Management — Update devices in the field via JSON-controlled OTA agent
• 🤖 AI Agent Framework — Add-on agents include:
  • ⚡ System Recovery Agent for fault detection and recovery
  • 🔧 Adapter Auto-Gen Agent to auto-generate configs for new devices (MQTT/OPC-UA)
  • 📦 Release Agent to run readiness tests and publish release notes
• 🧪 Vision Inspection Demos — ONNX model generator + camera inferencing pipeline
• 🔍 Repo Healthcheck — Lint and structure audit via tools/healthcheck.py

---

🤝 Sales + Technical Collaboration

This starter kit aligns with Sintrones’ efforts to:

• 🛠️ Support System Integrators and SMEs with demo-ready tools
• 🤝 Collaborate on R&D and Proof-of-Concepts
• 🌏 Promote industrial AI adoption across Thailand & SEA markets

Use it as a base to build your own PoC, integrate with IIoT, or contribute modules! this repo helps you accelerate time-to-demo and validate value at the edge.

---

🆓 vs 💼 Commercial Version

Feature	Open-Source Starter Kit	Commercial Offering
Real-time AI Inference (YOLO, etc.)	✅ Yes	✅ Yes
Dashboard UI (Streamlit/Grafana)	✅ Yes	✅ Yes
OTA Agent	✅ Yes	✅ Enhanced
Health Monitoring	✅ CLI Tool	✅ Web Dashboard
AI Agent Automation (Recovery, Adapter)	✅ Yes	✅ Advanced
Odoo / Cloud / AWS Integration	🟡 Manual	✅ Plug-in Ready
Hardware Acceleration Support	🟡 Generic	✅ Tuned Drivers
Long-term Support + SLA	❌	✅ Yes
Turnkey Packaging (VM/Image)	❌	✅ Yes

---

📦 Deployment Options

Mode	Description
Standalone	Fully offline dashboard & sensor integration
Edge-to-Cloud	MQTT to Odoo, AWS, or other IoT platforms
Vehicle AI	Add GPS/CANbus for on-road deployments

---

🎯 Use Cases

• 📦 Smart Logistics – Detect vehicles or goods, monitor temperature/vibration
• 🏭 Factory Automation – Visual inspection + machine health monitoring
• 🏙️ Smart Cities – Public space detection, traffic analytics, air quality

---

📚 Additional Resources

• 📘 Use Cases: Real-world Edge AI applications in factories, vehicles, and smart cities
• 🤝 Contributing Guide: How to get involved and contribute to this project

---

AI Agents Add-on

This repository integrates an AI Agents Add-on with three useful agents to enhance reliability, adaptability, and release workflows.

1. System Recovery Agent

• Purpose: Monitors MQTT heartbeat topics (e.g., factory/health/#).
• Behavior: If a device misses heartbeats for a configured timeout, it triggers recovery actions (e.g., restart services or notify operators).
• Usage:

  python -m src.agents.system_recovery_agent --config agents/system_recovery.yaml

2. Adapter Auto-Gen Agent

• Purpose: Inspects new devices and automatically generates adapter configuration snippets.
• Modes:
  • MQTT sniff mode: listens to wildcard topics and infers field mappings.
  • OPC UA browse mode: enumerates nodeIds and proposes mappings.
• Usage:

  # MQTT mode
  python -m src.agents.adapter_autogen_agent --mode mqtt --host localhost --topic factory/# --samples 30 --timeout 20
  
  # OPC UA mode
  python -m src.agents.adapter_autogen_agent --mode opcua --endpoint opc.tcp://192.168.10.20:4840

3. Release Agent

• Purpose: Automates readiness checks and drafts GitHub release notes.
• Behavior: Runs tools/healthcheck.py, verifies syntax & dependencies, and generates release notes into dist/release_notes.md.
• Usage:

  python -m src.agents.release_agent --tag v0.3.0 --notes "Adapters + Vision QA"

4) MQTT topics (default)

• factory/vision/detections – raw detections (per frame)
• factory/vision/events – filtered/decided events (if you wire through decision engine)

5) Troubleshooting

• Model not found: ensure models/defect_detector.onnx exists or pass an absolute path with --model.
• Unsupported IR version: upgrade onnxruntime or re-generate model with IR=10.
• No camera: use --video with a test clip.
• Broker connection: start Mosquitto locally or point to your broker in examples/vision_inspection/camera_infer.py (MQTT_HOST/PORT).

Requirements

Install additional dependencies with:

python -m pip install -r requirements-addon.txt

Outputs

• Recovery logs: console output
• Auto-generated configs: dist/config.autogen.yaml
• Release notes: dist/release_notes.md

For more details, see docs/AGENTS.md.

---

🛠️ Project Structure

sintrones-edge-ai-starter-kit/
├─- agents/                # Agent configs (e.g., system_recovery.yaml)
├── ai_models/             # YOLOv5 or OpenVINO model files
├── app/                   # Core dashboard + logic
│   └── main.py
├── configs/               # System & sensor configuration files
│   └─ config.yaml
├── dashboard/             # Streamlit and Grafana dashboard configs
├─- dist/                  # Auto-generated configs and release notes
├── docker/                # Dockerfile + docker-compose.yml
├── docs/                  # Wiring diagrams, ABOX-5220 architecture
│   └── index.md
│   └─ AGENTS.md           # Documentation for AI Agents
├── examples/              # Application-specific integration (vehicle, factory, city)
│   └─ vision_inspection/...
├─- models/
│   └─ defect_detector.onnx
├── ota/                   # OTA update agent and JSON control
├── sensor_drivers/        # CANbus, Modbus, GPIO, MQTT handlers
├─- src/
│   ├─ agents/             # AI Agents (system recovery, adapter autogen, release agent)
│   ├─ collector.py
│   ├─ batcher.py
│   ├─ cli.py
│   └─ decision_engine/
│      └─ engine.py
├─- tools/
│   └─ healthcheck.py       # Repo healthcheck tool
├─- requirements.txt
├─- requirements-addon.txt  # Dependencies for AI Agents
├── INSTALL.md
├── README.md
├── LICENSE
├── .gitignore

---

⚡ Getting Started

1. Clone the repository:

   git clone https://github.com/sintrones/edge-ai-starter-kit.git
   cd edge-ai-starter-kit

2. Install Python dependencies:

   pip install -r requirements.txt

3. Run the dashboard demo:

   python app/main.py

Vision Inspection Camera Publisher

This example publishes per-frame detections to MQTT for the collector to ingest. It supports a real ONNX model or a mock fallback.

1) Install dependencies

# Core
python -m pip install onnxruntime opencv-python paho-mqtt

# Apple Silicon (M1/M2/M3): use the silicon wheel
# python -m pip install onnxruntime-silicon opencv-python paho-mqtt

2) Prepare a model (optional)

If you already have an ONNX model (e.g., YOLO export), put it in models/defect_detector.onnx.

Otherwise, generate a tiny test model (always outputs one detection):

python models/onnx-model-generator/generate_dummy_onnx_with_onnx.py
# -> writes models/defect_detector.onnx

Note: If you see an error like Unsupported model IR version: 11, max supported IR version: 10, either upgrade onnxruntime (pip install --upgrade onnxruntime or onnxruntime-silicon) or regenerate the model with IR=10.

3) Run the camera publisher

Use a webcam:

python examples/vision_inspection/camera_infer.py \
  --model models/defect_detector.onnx --camera 0

Or a sample video:

python examples/vision_inspection/camera_infer.py \
  --model models/defect_detector.onnx --video path/to/sample.mp4

If you don’t have a model yet, run the mock-fallback script (publishes synthetic detections periodically):

python onnx-model-generator-ready/camera_infer_mock_fallback.py --camera 0
# or
python onnx-model-generator-ready/camera_infer_mock_fallback.py --video path/to/sample.mp4

# Collector should already be running to write JSONL
python -m src.cli collect --config configs/config.yaml

# Batch to Parquet
python -m src.cli batch --config configs/config.yaml

---

🧠 ONNX Model Troubleshooting

If you encounter errors like:

Unsupported model IR version: 11, max supported IR version: 10

You can either:

1. Upgrade your ONNX runtime:

   pip install --upgrade onnxruntime

2. Re-export the model with ir_version=10

---

⚡🔧📦 AI Agents Quick-Start

• ⚡ System Recovery Agent

  python -m src.agents.system_recovery_agent --config agents/system_recovery.yaml

• 🔧 Adapter Auto-Gen Agent

  # MQTT mode
  python -m src.agents.adapter_autogen_agent --mode mqtt --host localhost --topic factory/# --samples 30 --timeout 20
  
  # OPC UA mode
  python -m src.agents.adapter_autogen_agent --mode opcua --endpoint opc.tcp://192.168.10.20:4840

• 📦 Release Agent

  python -m src.agents.release_agent --tag v0.3.0 --notes "Adapters + Vision QA"

  Install add-on dependencies:

  pip install -r requirements-addon.txt

  Outputs include:

  • dist/config.autogen.yaml
  • dist/release_notes.md

  ---

MQTT Broker (Mosquitto) — Quick Run

Run locally:

# macOS (brew)
brew install mosquitto
brew services start mosquitto

# Ubuntu/Debian
sudo apt update && sudo apt install -y mosquitto mosquitto-clients
mosquitto -v  # foreground mode with verbose logs

Or via Docker:

docker run -it -p 1883:1883 -p 9001:9001 eclipse-mosquitto

Test:

mosquitto_sub -h localhost -t "test/topic"
mosquitto_pub -h localhost -t "test/topic" -m "hello"

---

🔍 Enhanced Vision Features (v1.1.0 Patch)

This patch includes additional AI/vision inspection modules for anomaly detection, logging, and RCA.

✅ New Capabilities

Feature	Module	Description
Anomaly Detection	anomaly/padim_infer.py	Basic visual anomaly scoring (extendable to PaDiM, DRAEM)
Frame Logger	logger/frame_logger.py	Save inspection snapshots with metadata
Root Cause Analysis (RCA)	clustering/image_cluster.py	Cluster visually similar defects using PCA + k-means
OTA Model Switch	ota/model_switcher.py + update_control.json	Swap ONNX models by version or line config

🔧 Usage Examples

Log a Frame and Metadata

from logger.frame_logger import save_frame_with_metadata
save_frame_with_metadata(image, {"line": "A1", "status": "PASS"})

Run Anomaly Detection

from anomaly.padim_infer import detect_anomalies
result = detect_anomalies(image)
print(result["is_anomaly"], result["anomaly_score"])

Switch Model via OTA Config

from ota.model_switcher import get_model_path_from_ota
model_path = get_model_path_from_ota()  # uses ota/update_control.json

Cluster Image Features (RCA)

from clustering.image_cluster import cluster_features
labels = cluster_features(feature_array, n_clusters=4)

These modules are designed to be extendable and can be linked with camera inference pipelines, OTA update agents, or retraining workflows.

---

🖥️ Streamlit Log Viewer

Use the built-in dashboard to browse visual inspection logs.

▶️ Launch Viewer

streamlit run dashboard/log_viewer.py

🔍 What It Does

• Displays image logs from logs/*.jpg
• Shows associated metadata from logs/*.json
• Sidebar shows count of PASS/FAIL units

---

🧪 Running Tests

Run all tests using:

pytest tests/

Tests include:

• Logger: saves annotated frame + JSON
• OTA model switch: reads ONNX path from control JSON

---

🧪 Automated Testing & CI

This project includes a growing suite of pytest-based unit tests found in the /tests folder.

Test Coverage Areas:

• logger.py — Logs each inference and frame snapshot
• ota/model_switcher.py — OTA JSON-based model switch controller
• src/agents/system_recovery_agent.py — Simple heartbeat-based recovery agent
• dashboard/log_viewer.py — Streamlit app to view inference logs and anomaly images

✅ GitHub Actions CI

GitHub Actions automatically runs tests on every push or pull request to main.
The test workflow includes:

• Python 3.10 setup
• Dependency install (requirements.txt, requirements-addon.txt)
• CI environment with PYTHONPATH for clean imports
• Full pytest run on /tests

See .github/workflows/python-ci.yml for the CI config.

---

📢 Community & Contact

• Website
• LinkedIn
• Edge AI Community (Coming soon)

📬 Want a hardware demo kit? Contact Sintrones

---

📄 License

MIT License — open for research, testing, and pilot deployment.