IoT meets AI: edge AI, TinyML, anomaly detection and AI agents that consume and act on MQTT data streams.
Letting a language model switch real equipment is a control problem, not a prompt problem. Learn how to put a command gateway with allowlists, limits, confirmations and an audit trail between the AI and your MQTT devices.
Learn how to turn streams of MQTT sensor readings into readable shift reports and anomaly explanations by computing statistics in code and letting an LLM write only the narrative.
A hands-on introduction to TensorFlow Lite for Microcontrollers: the model, op resolver, tensor arena and interpreter explained, a complete example on an ESP32-S3 board, and a serial-to-MQTT bridge that works for any board.
A practical guide to detecting anomalies in sensor streams: which methods fit which signals, whether to run them on the device, the edge or in the cloud, and how MQTT ties detectors and alerts together.
A realistic look at where generative AI helps on the factory floor, from shift summaries to troubleshooting assistants, and how to ground every answer in live machine data streamed over MQTT.
What sensor fusion is, how complementary and Kalman filters combine noisy sensors, when to fuse at the edge versus in the cloud, and working ESP32 and Python examples that publish fused values over MQTT.
Build an edge vision pipeline on an NVIDIA Jetson: run GPU-accelerated detection, publish compact JSON results and device health over MQTT with TLS, and run it as a service that survives reboots.
A hands-on tutorial on IoT time series forecasting: capture sensor data from MQTT, resample it, compare a statistical model with a machine learning model, backtest honestly and publish forecasts back to the broker.
A practical comparison of running IoT models on devices versus in the cloud, with a decision checklist and a hybrid architecture where edge results flow over MQTT to cloud models.
How remaining useful life prediction works, which model families exist, what data history each needs from MQTT-connected assets, and how to publish RUL estimates with uncertainty.
A complete tutorial that streams simulated sensor data over MQTT, scores every reading with rolling statistics in Python and publishes debounced anomaly events to an alerts topic.
Run a tiny person detection model directly on an ESP32-CAM, publish a confidence score and debounced alerts over MQTT, and keep images on the device unless you explicitly need them.
Train an Isolation Forest on historical multivariate sensor data with scikit-learn, then load it in a Python service that scores live MQTT readings and publishes anomaly events.
Run a YOLO model on a webcam or RTSP stream in Python, turn each frame's detections into compact JSON, and publish counts and alerts over MQTT so any dashboard or automation can react.
Step-by-step tutorial for a Model Context Protocol server that subscribes to MQTT topics, caches the latest values and exposes read and command tools to AI assistants.
A practical use case for energy load forecasting: collecting meter data over MQTT, turning energy counters into demand, adding weather and calendar features, and publishing day-ahead forecasts for buildings and plants.
Take a trained Edge Impulse model off the Studio and into a real system: export it as an Arduino library or a Linux .eim file, run inference on the device and publish every prediction over MQTT.
Federated learning lets IoT devices improve a shared model while their raw data stays local. This explainer covers how it works, where it fits, its pitfalls, and how an MQTT broker can carry model updates between devices and a coordinator.
A practical comparison of edge AI hardware from microcontrollers to GPU modules, with a decision checklist and a benchmark script that publishes your own model's latency over MQTT so you can compare boards fairly.
A hands-on tutorial that turns requests like "dim the kitchen lights to 30 percent" into validated MQTT commands, using an LLM with function calling, a device registry and a strict validation layer.
A complete TinyML ESP32 walkthrough: train a vibration classifier in Python, convert it to an int8 TensorFlow Lite model, run it on an ESP32 with esp-tflite-micro and publish classification results to an MQTT broker.
A practical guide to shipping machine learning models to fleets of edge devices, using MQTT retained messages to announce new versions and collect rollout status from every device.
Machines often sound wrong before they fail. Learn how acoustic anomaly detection works, which models fit edge hardware, and how to publish compact anomaly scores over MQTT instead of streaming raw audio.
How to capture machine events and operator labels over MQTT alongside telemetry, define a label schema, convert events into labeled intervals and join them with sensor windows for supervised learning.
A practical guide to retrieval-augmented generation for factories: index manuals and procedures, pull current values from MQTT topics and give technicians answers grounded in both, with citations.
Why publish/subscribe suits multi-agent AI systems, and the topic conventions, MQTT 5 features and safety rules that let agents discover each other and exchange messages through a broker.
Edge AI means running machine learning inference on or near the device that collects the data. This explainer covers how it works, the hardware tiers, and why edge devices should publish results rather than raw data.
A practical tutorial for running Ollama on an IoT gateway, subscribing to MQTT telemetry and publishing structured, AI-generated insights back to the broker without sending data to an external AI service.
Learn what Frigate NVR publishes over MQTT, how to read the frigate/events payload and how to forward only the detections that matter to a cloud broker for remote notifications.
A practical guide to combining AI and IoT: where models should run, how sensor data reaches them, how to detect anomalies in real time and how to let LLM agents act on devices safely.
Create an account, get your broker credentials and connect your first device — no servers to install, patch or scale.
