A major challenge for many companies today is managing the vast amounts of fragmented raw data generated by connected devices across different generations, hardware revisions, and vendor protocols. Traditional platforms require rigid, hand-coded adapter pipelines for every single variation.
Imagine a future where you never have to write another rigid, custom parser for a new device on your network. By combining the universal ingestion of the Cumulocity MQTT Service with the power of Agentic AI, we are aiming to shift your team’s focus from how to parse data to what outcomes you can achieve with it.
While the Data Preparation application is currently in Public Preview, we want to share a hands-on look at how this AI-powered approach handles real-world payloads today. To demonstrate how natural language prompting can simplify incoming MQTT transformations, we’ve put the Data Preparation Expert Agent to the test across 14 enterprise IoT data formats (delivered over MQTT):
- Standard Telemetry: MQTT JSON, SensorThings, Azure CSV, CSV Flat File
- Industrial & Utilities: Sparkplug B, OPC UA, Modbus, LwM2M (CBOR/JSON/TLV)
- Compact & Binary: Protobuf, CBOR, LoRaWAN (ChirpStack / Actility Base64)
Below are 3 practical walkthroughs showing what’s working right now, along with an invitation to test your own data formats alongside us.
1. The “Hello World” Example: Standard JSON payload
Let’s start with the basics. You have a modern sensor sending clean, well-formatted JSON over MQTT, in this case a simple tracking use-case for a refrigeration container.
{"timestamp": "2026-03-26T16:10:00.000Z", "position": {"lat": 31.50000, "lng": 35.40000, "alt": 420.0}, "temperature": "-4F"}
I can then simply ask the AI assistant, if he would not already propose it, to “Convert this into a Cumulocity measurement”
The Expert Agent then interprets the data, with his understanding of Cumulocity data model and proposes the most appropriate mapping - in this case to a Position event and a temperature measurement.
The result is a nicely written and explainable piece of javascript code: see explanation of what it does on the left (as part of the Expert Agent’s initial response - and which can be redone at any point in time by clicking the “explain code” button) as well as well-documented code on the right.
The proof of the pudding is in the eating, so let’s validate the working of the Smart Function by clicking “Run tests”; which nicely results in the expected mapping into a temperature measurement and a location update via a c8y_Position fragment
2. The Industrial Standard: Sparkplug B
In manufacturing and industrial operations, Sparkplug B is a widely adopted specification for using MQTT in a state-aware, interoperable way. Its payloads are encoded with Protocol Buffers: compact and structured, but not directly readable without the appropriate decoding logic.
PAYLOAD = bytes.fromhex("08e0aa9ebbd733121f0a0b54656d706572617475726518e0aa9ebbd733200a690000000000205240121c0a0848756d696469747918e0aa9ebbd733200a6933333333335350401800”)
The Cumulocity Data Preparation Agent recognizes Sparkplug B and other structured payload formats from the incoming data. It can validate that a sample matches the expected Sparkplug B Protobuf structure, interpret the contained metrics and metadata, and generate a native transformation rule that maps them into the Cumulocity data model.
After interpreting the data, The Expert Agent then explains how it creates the resulting transformation function:
Again, running the test reveals a nicely formatted Cumulocity measurement containing temperature and humidity fragments and series.
As such, instead of spending time writing and maintaining custom transformation logic for industrial metrics, teams can focus on defining the outcomes they want to achieve while Data Preparation assists with generating the required rules.
3. The Compact Binary: LoRaWAN Payloads
In low-power networks such as LoRaWAN, sensors transmit compact binary payloads over the air to conserve battery life and airtime. LoRaWAN Network Servers (LNS) such as ChirpStack and Actility ThingPark then forward each message to MQTT, together with device and network metadata. Depending on how the LNS is configured, it can publish the telemetry to MQTT in one of two forms:
-
Pre-decoded JSON: When a device payload codec is active, the LNS decodes the sensor payload into structured JSON. The AI Agent can map the named telemetry fields directly to Cumulocity, just as in Example 1.
-
Raw Base64-encoded payload: When no codec is configured, the LNS forwards the original device payload as a Base64-encoded string. The AI Agent must first decode the device-specific payload before mapping its measurements to Cumulocity.
In the latter case, instead of writing custom byte-parsing scripts, you simply supply the sensor manufacturer’s standard JavaScript codec (eg. decodeUplink() ) directly to the AI Agent (or simply describe the payload layout in natural language) and the agent generates the decoding and mapping logic.
Given a raw Base64 string, the AI automatically interprets the bitwise logic:
- extracting the signed 16-bit big-endian integer for temperature (22.5°C)
- extracting the byte for humidity (45%)
and generates a native, performant transformation rule in seconds. No custom middleware, no platform plugins, and no manual byte-wrangling required.
Ready to Accelerate Data Ingestion?
The Data Preparation application is available now in Public Preview across all Cumulocity SaaS instances. To test it with your own payloads:
-
Enable the Feature: In the Administration application, click your user profile (top right corner), select Manage preview features, and enable Data Preparation.
-
Activate the AI: Ensure the AI Agent Manager is configured on your tenant to unlock natural language assistant capabilities.
Stop letting payload formatting slow down your IoT roadmap.
Bring your raw device data, and let the Cumulocity Data Preparation Expert Agent make sense of it.
What’s Your Data Format?
We’ve validated the Data Preparation Expert Agent against 14 enterprise IoT data formats delivered over MQTT —including JSON payload, Sparkplug B, OPC UA, LwM2M, Modbus, Protobuf, CBOR, and LoRaWAN.
Now we want to hear from you: Is there a custom data format you don’t see on our list? Bring us your binary payloads, legacy hex dumps, or proprietary MQTT message structures, and help us validate format #15!
