Telemetry pipelines from
device to decision
We build the layer between physical assets and the people responsible for them: authenticated device connectivity, ingestion that survives bad links, stream processing with real alarm semantics and storage that stays queryable years later.
Platform capabilities
Every layer below has been built in production for energy storage, charging and fleet telemetry — not sketched for a proposal.
Device connectivity
MQTT, HTTP and WebSocket transports, plus protocol adapters for gateways and controllers that speak something else. New device families are added as adapters, not as forks of the core.
Device identity and access
Per-device credentials, token-based authorisation and scoped topics. A compromised gateway cannot read or publish outside its own assets.
Fault-tolerant ingestion
Buffering, retry with backoff and deduplication on both edge and server side. Devices on flaky links backfill instead of losing their history.
Stream processing
Filtering, aggregation, enrichment and anomaly detection on the live stream. Threshold and rule-based automation fires within seconds of the reading that triggered it.
Time-series storage
Retention tiers, downsampling and continuous aggregates so a year of history stays queryable without keeping every raw sample hot.
Operator visibility
Fleet and asset overviews with status, geo-location and session timelines, custom widgets and drill-down, scoped by role for operations, support and management.
How a reading travels
Six stages between a sensor and an operator. Most IoT platforms fail at the boundaries between them, so that is where we spend the design effort.
Edge and gateway
Device firmware or gateway agents publish structured telemetry. We define the payload schema and topic layout before anything else is built.
Ingestion
Broker or HTTP endpoint with authentication, rate limiting, buffering and deduplication. Malformed payloads are quarantined, never silently dropped.
Normalisation
Units, timestamps and device metadata are reconciled into one canonical model so downstream code never has to know the vendor.
Processing and rules
Aggregation windows, anomaly detection and alarm rules evaluated on the live stream, with severity tiers and source attribution on every event.
Storage
Hot time-series for dashboards, rolled-up aggregates for trends, cold archive for compliance. Query patterns decide the layout.
Consumption
Operator dashboards, alerting channels, BI extracts and back-office integrations all read the same normalised data.
What runs under the hood
Deployable to cloud, on-premise or hybrid infrastructure without rewriting the application layer.
- MQTT
- HTTP
- WebSocket
- Protocol adapters
- Node.js
- Kafka
- RabbitMQ
- Rule engine
- TimescaleDB
- PostgreSQL
- ClickHouse
- Redis
- Docker
- Kubernetes
- OpenTelemetry
- Structured logging
Connected expertise
Dashboards→
The operator interfaces that turn this telemetry into something a team can act on.
BESS monitoring→
The same pipeline applied to battery storage and PV inverters over Modbus, IEC 61850 and SunSpec.
Cloud & on-prem→
Where the platform runs: cloud, on-premise or hybrid, with the same deployment model.
Designing a telemetry platform?
Tell us the device families, message rate and retention you need. We will outline an ingestion, processing and storage architecture that scales from pilot to production without a rewrite.