Imagine constructing a system that monitors industrial equipment, collects sensor readings, detects unusual behavior and alerts a maintenance team. Writing the application is only one part of the work. The complete system may also require hardware integration, reliable data pipelines, networking, and machine learning.
An important question is then, do software companies build AI tools and IoT systems, or do these projects require specialized engineering teams?
The answer is not a clear-cut, but many software companies indeed develop these solutions. A complex system often requires expertise from several areas.
Who builds AI tools? A closer look
An AI development often extends beyond merely plugging an application together with a model. Depending on the use case, a team may need
software developers to build applications, APIs, and user interfaces,
machine learning engineers to integrated, evaluate, deploy, and maintain the models,
data engineers to build reliable pipeline for delivering the data,
and cloud and infrastructure engineers to manage the deployment, scaling, and monitoring.
A document summarization application may for example solely use an existing AI model through an API, whereas a predictive maintenance system may require historical equipment data, feature engineering, model evaluation, and integration with operational software.
Building a model from scratch is sometimes not beneficial, and for many projects the main engineering challenge consists of utilizing an existing model into a reliable application.
Who builds IoT systems? A closer look
The Internet of Things (IoT) concerns interconnected physical devices, allowing them to collect and exchange information
An IoT project may involve
embedded developers who write firmware for microcontrollers and connected devices,
IoT engineers who configure device connectivity and communication protocols,
hardware engineers who construct sensors, circuits, and other electronic components,
backend developers that receive and process the data from devices,
and security engineers who safeguard devices, networks, and information.
The IoT technology stack is highly varied depending on the project, and can span from the MQTT protocol for messaging between devices and services, HTTP-based APIs for interacting with an application backend, databases for storing device telemetry, edge or cloud-based processing of the information.
The choices made in the technology stack depend on latency requirements, connectivity, power constraints, security, and operating costs.
How do AI and IoT work together? A closer look
When AI is combined with IoT, the result is often referred to as AIoT, or Artificial Intelligence of Things. For example, an industrial monitoring system may have the following elements
1. Sensor data collection — sensors that keep track of equipment conditions,
2. Information communication — a device or gateway that forwards the information to a processing system,
3. Data processing — validate and enrich the data, and store the valuable readings in a database,
4. AI inference — use a model to identify unusual patterns,
5. Operational response — the application acts upon the recommendations from the model, for example, send an maintenance alert.
Each of these steps introduces their own engineering challenge. The sensor data may be noisy, the communications interrupted, and an AI model produce false positives. A production system therefore requires data validation, monitoring, error handling, and an evaluation system for the model. Predictions should also be weighted according to their impact and risk before initiating an action request — in industrial settings such manual review may be prudent before scheduling a maintenance.
Should everything always run in the cloud? A closer look
Depending on requirements, a distributed processing architecture may be preferable, especially when devices need to respond to events with low latency a constraint the hardware may impose.
Edge computing keeps data processing close to the devices, and can be utilized to reduce communication overhead, or avoid bandwidth-intensive raw sensor data to be sent to another processing system.
Cloud-based processing offers powerful computational resources and storage, but have communication and security requirements of their own.
A hybrid architecture allows a system to both perform computations on the device or at the edge, while also storing data in the cloud for long-term storage, or more involved processing later. An edge gateway may for example filter sensor readings, whereas the cloud keeps a copy of the data for later big-data analysis.
The best architecture is determined by the requirements of the system at hand — not necessarily running everything in the cloud is often a sensible decision.
Do businesses need separate AI and IoT teams? A closer look
It depends on what is being built, but a small AI application may only require some software and AI expertise, whereas a connected device product may benefit from embedded engineering, hardware integration, and backend developers. More involved industrial platforms may require all of the above, including cybersecurity and systems engineering capabilities. Some companies specialize in isolated technical disciplines, whereas others seek to combine several. An example is (https://apertureventurestudio.com/), that focus on developing technology ventures at the intersection of AI and IoT for industrial applications.
The important consideration is understanding the problem, the technical requirements, and how the system will operate to fulfill an objective.
What should developers consider before starting an AIoT project? A closer look
Before choosing a framework, model, or hardware platform it is often prudent to outline the problem, and system requirements.
The following points are useful to assess before developing any technical solution:
What information must the system collect?
What sensors or devices can provide this information?
Is AI necessary, or can conventional software solve the problem?
Where should data be processed: on the device, edge, or cloud?
How will the system handle missing data and network interruptions?
How will security, monitoring, updates, maintenance, and scalability work?
How will the team evaluate whether the solution has value?
These considerations help avoid overly ambitious technical solutions, and provide clarity in design decisions.
Final thoughts
AI tools and IoT systems are built by software developers, AI specialists, embedded engineers, hardware professionals, and infrastructure teams. Software offers the applications and integrations, IoT links digital systems to the physical world, and AI provides data interpretation when intelligent analysis is beneficial. The challenge is that all these elements must work together reliably. For aspiring developers venturing into these fields, understanding the complete system is arguably more important than knowing a particular programming language, framework, or protocol.
Discussion: When designing an industrial AIoT system, do you prioritize data collection, processing, or AI inference first? What trade offs would you make?
Top comments (0)