Latency too high for cloud processing
Real-time detection or control cannot tolerate round trips to distant data centres.
01 / Advance Technology
Internet of Things and Edge AI
Some decisions cannot wait for data to travel to the cloud and back. Cameras on a production line, sensors on remote equipment and devices with patchy connectivity need intelligence close to where data is produced. We design connected systems where devices, edge computing and cloud each do the right part of the work.
At a glance
- Service
- Internet of Things and Edge AI
- Pillar
- Advance Technology
- Typical starting point
- Use case, device and connectivity assessment
- Typical first phase
- 3 to 5 weeks for architecture; pilots in 8 to 16 weeks
- Key technologies
- Internet of Things, Edge AI, Computer Vision
Durations depend on scope, data readiness and integration needs.
What is IoT and Edge AI?
Short answer
Internet of Things (IoT) and edge AI is the design of connected systems in which sensors and devices collect physical-world data, and AI models run on the device or on nearby edge hardware to analyse that data locally, sending only relevant results or summaries to cloud platforms for storage, fleet management and further analysis.
An IoT system has several layers: sensors and devices, connectivity, edge gateways, cloud platforms and applications. Edge AI adds models that run on constrained hardware, which introduces its own engineering challenges around model size, power consumption, updates and security.
We decide where each piece of processing should happen based on latency, bandwidth, privacy, reliability and cost. We then design the hardware selection, firmware and software, model optimisation, secure device management and the cloud services that tie the fleet together.
Business problems
What problems does IoT and Edge AI solve?
These are the situations where organisations most often find this service worthwhile.
Limited or costly connectivity
Remote sites, vehicles or farms cannot stream raw data continuously.
Privacy constraints on raw data
Video or audio should not leave the premises, but insights are still needed.
Unused equipment data
Machines and facilities produce data that nobody collects or analyses.
Manual monitoring of dispersed assets
Staff travel to check equipment that could report its own condition.
Insecure device fleets
Connected devices are deployed without proper identity, encryption or update mechanisms.
Use cases
Where can IoT and Edge AI be used?
Representative applications. The right use case depends on your processes, data and constraints.
| Area | Example application |
|---|---|
| Manufacturing | Smart cameras for defect detection and vibration sensors for machine health. |
| Agriculture | Soil and weather sensors, pest detection traps and irrigation control. |
| Smart buildings | Occupancy sensing, energy optimisation and equipment monitoring. |
| Logistics | Cold chain monitoring, vehicle telematics and warehouse vision. |
| Utilities and infrastructure | Remote monitoring of pumps, transformers, pipelines and structures. |
| Retail | Shelf monitoring and footfall analytics processed on-site. |
Who should consider it
- Manufacturers and industrial operators
- Infrastructure and utility organisations
- Agritech and logistics companies
- Product companies building connected devices
How AARHIT approaches it
How we deliver IoT and Edge AI
Each stage ends with something you can review, so decisions are made on evidence rather than on promises.
Stage 1 of 6
Use case and environment assessment
We define what needs to be detected or measured, where, how quickly and under what physical and connectivity conditions.
Output: requirements covering latency, power, environment and connectivity.
Stage 2 of 6
Architecture and hardware selection
We decide the split between device, edge and cloud, and select sensors, compute modules and connectivity.
Output: system architecture and bill of materials options.
Stage 3 of 6
Data collection and model development
Data is collected in real conditions, labelled and used to develop models suited to the target hardware.
Output: trained models and performance benchmarks on target devices.
Stage 4 of 6
Edge software and optimisation
Models are quantised or compiled for edge accelerators, and device software handles inference, buffering and communication.
Output: edge application with measured latency and power use.
Stage 5 of 6
Cloud platform and fleet management
Secure provisioning, telemetry, remote configuration, over-the-air updates and dashboards are implemented.
Output: device management and data platform.
Stage 6 of 6
Field pilot and scale-up
Devices are deployed at pilot sites, performance is validated and processes for installation and support are defined.
Output: pilot results and a rollout plan.
Technology
What technologies are involved?
We choose tools on fit, cost, security and your existing estate. These are typical options, not fixed commitments.
Learn more about the underlying technologies
| Layer | Typical options |
|---|---|
| Edge hardware | NVIDIA Jetson, Raspberry Pi and industrial PCs, Coral and Hailo accelerators, ESP32 and STM32 microcontrollers |
| Connectivity | Wi-Fi, 4G and 5G, LoRaWAN, NB-IoT, Bluetooth Low Energy, industrial Ethernet |
| Protocols | MQTT, OPC UA, Modbus, HTTP and CoAP |
| Edge AI tooling | TensorFlow Lite, ONNX Runtime, TensorRT, OpenVINO, model quantisation |
| IoT platforms | AWS IoT, Azure IoT Hub, open-source brokers and device management |
| Security | Device identity, secure boot, encrypted communication and signed updates |
Outcomes
Potential business outcomes
Targets are agreed with you during scoping and measured against a baseline. We do not quote improvement figures before we have seen your process and data.
- Real-time detection and response where it matters
- Lower bandwidth and cloud costs through local processing
- Better privacy by keeping raw data on-site
- Remote visibility of dispersed assets
- Secure, updatable device fleets
How long does implementation generally take?
Typical timeline
Architecture and hardware selection typically take 3 to 5 weeks. Field pilots, including data collection and model optimisation, usually run 8 to 16 weeks. Scale-up depends on hardware procurement and installation logistics.
What determines complexity?
- Environmental conditions and ruggedisation needs
- Power availability and battery constraints
- Model accuracy and latency requirements
- Fleet size and geographic spread
- Integration with existing control or enterprise systems
Implementation
Implementation considerations and risks
Most difficulties in this kind of work are predictable. Planning for them early keeps delivery on track.
Collect data where devices will run
Models trained on clean lab data often fail under real lighting, noise, dust or weather. Field data is essential.
Plan updates from day one
Devices in the field need secure over-the-air updates for software and models. Adding this later is difficult.
Consider installation and maintenance
Mounting, power, cleaning and replacement procedures affect total cost and reliability as much as the technology.
Check certification needs
Radio, electrical and industry certifications may be required before commercial deployment.
| Risk to consider | How we manage it |
|---|---|
| Model performance drops in the field | Field data collection, on-device monitoring and remote model updates. |
| Device security vulnerabilities | Unique device identities, secure boot, encryption and patch processes. |
| Hardware supply issues | Alternative component options and early procurement planning. |
| High support costs at scale | Remote diagnostics, standard installation kits and clear maintenance routines. |
FAQ
Questions about IoT and Edge AI
What does an IoT and edge AI engagement include?
Typically use case and site assessment, device, connectivity and hardware selection, field data collection, model optimisation for target devices, secure fleet management with over-the-air updates, and a field pilot before rollout.
When should AI run at the edge rather than in the cloud?
Edge processing makes sense when decisions must be made in milliseconds, connectivity is unreliable or expensive, raw data is sensitive, or data volumes such as video are too large to transmit continuously.
Can edge devices run large language models?
Small language models can run on capable edge hardware for specific tasks. Larger models usually run in the cloud or on local servers. We choose based on the task, latency and hardware budget.
How are models updated on devices in the field?
Through secure over-the-air update mechanisms that verify signed software and model packages, roll out gradually and can roll back if problems appear.
Do you design custom hardware?
We specify and integrate hardware, often using proven modules, and work with electronics design and manufacturing partners when custom boards or enclosures are needed.
Next step
Considering IoT and Edge AI for your organisation?
Share the process, product or decision you have in mind. We will ask the right questions, outline options and suggest a sensible first step.