On-device intelligence for mesh sensor networks.

Train, validate, and deploy ML models that run directly on resource-constrained BLE mesh nodes — from data collection through edge inference, without cloud dependency.

See the Pipeline Explore Hardware

Pipeline

From raw signals to deployed models.

01

Collect

Hardware sensors stream IMU, audio, and environmental data at configurable rates.

02

Preprocess

Edge nodes filter, downsample, and extract features before transmission.

03

Train

Versioned experiments across TensorFlow, PyTorch, and ONNX with auto-tuning.

04

Validate

Holdout sets, A/B comparisons, and drift detection before promotion.

05

Deploy

OTA model push to Cortex-M targets with rollback and fleet monitoring.

Capabilities

Built for the edge, not the cloud.

Every stage of the ML lifecycle runs on or close to the hardware — minimizing latency, bandwidth, and cloud costs.

Data Collection

  • IMU, BLE RSSI, PDM microphone, temperature
  • Configurable sampling rates per sensor
  • Edge preprocessing reduces bandwidth 10x

Model Training

  • TensorFlow, PyTorch, ONNX runtime support
  • Versioned experiment tracking
  • Hyperparameter auto-tuning pipelines

Continuous Learning

  • Privacy-aware on-device telemetry
  • Drift detection with automated retraining
  • Rolling model updates without downtime

Edge Inference

  • Streaming analytics on Cortex-M MCUs
  • Sub-10ms classification latency
  • Smart alerts with configurable thresholds

Integrations

Connects to what you already use.

Home Assistant

Native MQTT integration for automations and dashboards.

Grafana

Time-series visualization with pre-built sensor dashboards.

REST API

Full CRUD for models, devices, and inference results.

WebSocket

Real-time streaming of inference events and alerts.

Webhooks

Signed payloads to any endpoint on classification events.

API

Signed webhooks on every event.

Every classification, alert, and model update triggers a signed webhook to your endpoint. Integrate with Grafana, Home Assistant, PagerDuty, or your own backend.

Sample payload

{
  "event_id": "evt_8f2a...",
  "type": "classification",
  "room": "living_room",
  "label": "motion_detected",
  "confidence": 0.94,
  "model_version": "v2.1.3",
  "timestamp": "2026-04-01T10:30:00Z"
}

Test Webhook Delivery

Simulate a classification event and see the webhook payload delivered in real time.

Ready to add intelligence to your mesh?

Start with the hardware, then layer on AI models as your deployment grows.

Get Early Accessdrewscreations.com