Tech talk · May 2026
From Sensor to Dashboard, a $10 chip to a live dashboard.
A real-time IoT pipeline built end to end with the same patterns as a production data stack, only smaller, cheaper and visible from the sensor to the screen.
Big Data Madrid · tech talk #4
Demo: real-time IoT pipeline, end to end
The problem
Production data stacks hide their plumbing: brokers, time-series stores and push channels sit behind managed services and dashboards.
The cheapest way to feel a real streaming pipeline is to build one you can hold: a sensor node, a broker and a database on one Raspberry Pi, with no cloud bill.
For
Data engineers, and anyone curious about streaming pipelines
Stack
What the talk covers
- 01
Five hops
Pico W with a DHT22 sensor → EMQX (MQTT) → TimescaleDB → FastAPI over WebSocket → the browser. About 150 lines of Python for the whole stack.
- 02
MQTT vs Kafka
A Kafka producer needs hundreds of megabytes of RAM; an MQTT client runs in kilobytes. Neither is better: the right protocol lives at the right layer.
- 03
A real time-series store
TimescaleDB hypertables on PostgreSQL: time-partitioned automatically, durable and replayable.
- 04
Push, not poll
PostgreSQL LISTEN/NOTIFY plus a WebSocket: the database is both the source of truth and the fan-out point, with zero polling.
- 05
The seams
Where this scales: swap the sensor for Kafka producers, TimescaleDB for ClickHouse, NOTIFY for Debezium-style change data capture. Same shape.
Key takeaways
- —Same patterns, different physics: everything here is something data engineers already know, at a different scale.
- —The MQTT topic is your schema: device, metric and sensor in one string.
- —PostgreSQL LISTEN/NOTIFY is a zero-cost pub/sub bus for live interfaces.
- —Protocol choice is a constraints choice: memory, network and durability decide it.
Want this in your stack?
A 30-minute decision call turns this pattern into a plan for your team, or book the talk for them.