AI
Manchester Edge AI Lab — On-Device AI & Embedded Systems

On-device AI
that runs when
the network doesn't.
Edge Intelligence. Shipped.

We design and ship on-device AI and embedded AI systems — local inference pipelines, computer vision, and edge intelligence solutions that run on your hardware, under your control.
Low latency. No cloud dependency. No data egress. From the silicon itself through to the app your team uses in the field.

Jetson Orin · Raspberry Pi 5 · Hailo-8 · Coral TPU · Rockchip RK3588 · NXP i.MX 9x · TensorRT · ONNX Runtime · TFLite · YOLOv8 · On-Device ML · Local Inference · EdgeOps Observability · Rust Systems · AWS Cloud · Made Smarter NW ·
Jetson Orin · Raspberry Pi 5 · Hailo-8 · Coral TPU · Rockchip RK3588 · NXP i.MX 9x · TensorRT · ONNX Runtime · TFLite · YOLOv8 · On-Device ML · Local Inference · EdgeOps Observability · Rust Systems · AWS Cloud · Made Smarter NW ·
Jetson
Orin / AGX / NX
Raspberry Pi
Pi 5 / CM4
Hailo
Hailo-8 / M.2
Coral
Edge TPU / Dev Board
NXP i.MX
i.MX 9x / eIQ NPU
Rockchip
RK3588 / RK3568
The Problem
01

Businesses have the vision for on-device AI and edge intelligence. They can't find anyone who can actually build it.

You have a product idea that needs local inference on embedded hardware — real-time object detection, autonomous operation without cloud connectivity, or data-privacy-first AI that never leaves the building. Finding a team who can deliver across the full stack is genuinely difficult.

Most consultancies handle the cloud layer or the model layer — not the embedded systems, custom Linux BSP, and low-latency inference pipeline underneath. That's the gap Manchester Edge AI Lab exists to close.

For regulated environments: regulated data doesn't mean no AI. It means on-premises AI — deployed locally on your hardware, under your control, with zero data egress and full GDPR compliance.

01 — What we build

What We
Actually Build

Custom Embedded
Linux Systems
Custom Linux builds from BSP integration upward. Device drivers, real-time configuration, hardened production system images, and bespoke display pipelines. Jetson-first, same depth on every supported edge computing platform. The foundation on which reliable on-device AI runs.
JetPack 6.xRPi OSYoctoBuildrootDevice TreesU-Boot
Edge AI
Local Inference Pipelines
High-performance local inference optimised for the silicon you're deploying on. TensorRT on Jetson for GPU-accelerated on-device ML. ONNX Runtime and TFLite (LiteRT) on Raspberry Pi and ARM targets. Hailo SDK for dedicated NPU hardware. FP16 and INT8 — low latency numbers on real hardware, not benchmark theatre.
TensorRTONNX RuntimeTFLite / LiteRTHailo SDKCUDARust
Computer Vision
& Object Detection
People counting, object detection, stereo depth, occupancy analytics, and industrial quality inspection. Hardware-synchronised camera pipelines, ISP tuning, GStreamer integration. Neural network inference from raw sensor output to production-ready results — on the device, at the edge.
YOLOv8 / v11Stereo DepthCSI / USBGStreamerOpenCV
Local AI for
Regulated Environments
Open-weight language models on your premises. No cloud, no data egress, no compliance risk. Llama, Mistral, Phi and Whisper on RTX workstations or Jetson Orin — connected to your systems via Conduo. For legal, healthcare, and manufacturing teams who need AI without sending sensitive data to third-party cloud infrastructure. GDPR-safe, by design.
Llama / MistralWhisperConduoOn-PremisesGDPR-safeData Privacy
Find out more →
Cloud &
Mobile Applications
Fleet telemetry, OTA updates, model versioning, and remote diagnostics in the cloud. Bandwidth-optimised data pipelines that transmit only what matters — processed edge intelligence, not raw sensor streams. Native iOS apps surfacing on-device AI insights for field operators in real time. Hardware to hand.
AWS IoT CoreGoSwift / SwiftUIOTA UpdatesMQTT
EdgeOps &
Observability
Runtime monitoring, latency profiling, anomaly detection, and remote diagnostics for deployed edge AI fleets. Fog computing connectivity between distributed inference nodes and the cloud. Powered by Conduo — our open-core EdgeOps platform built for the realities of autonomous edge intelligence deployments.
ConduoPrometheusGrafanagRPCFleet Mgmt
02 — Real numbers

Benchmarks,
Not Claims

3.51ms
Local inference latency
YOLOv8n object detection on Jetson Orin Nano 8GB. TensorRT FP16. 284.9 FPS throughput. On-device, autonomous, no cloud round-trip.
trtexec · JetPack 6.2 · CUDA 12.6
25+
Years embedded systems
Motorola, JLR, Intel, NXP. Linux kernel, device drivers, Khronos/Vulkan contributor, native iOS, AWS serverless. BSP upward. Every layer. Open source throughout.
NVIDIA Connect · Linux Foundation contributor
Multiple platforms
Edge AI hardware
Jetson, Raspberry Pi, Hailo, Coral, NXP i.MX, Rockchip, Ri. We recommend the right edge computing hardware for your brief and deliver across the full stack — from embedded AI to cloud.
INT8/FP16
Quantised model deployment
Post-training quantisation and TinyML techniques for power-constrained deployments. The same model accuracy — running on hardware that fits in your product.
TensorRT · ONNX Runtime · TFLite · Hailo SDK
Live benchmark · local inference

Real hardware.
Real numbers.

Every project starts with a benchmark on your target edge AI hardware. We don't estimate local inference performance — we measure it. Low latency. No cloud round-trip. Autonomous operation from day one.

orin-nano — jetpack 6.2 — cuda 12.6
# YOLOv8n on-device object detection · TensorRT FP16
trtexec --loadEngine=yolov8n_fp16.engine \
--iterations=1000 --warmUp=500
# ── local inference results ──────────────────────
mean latency  3.51 ms
throughput  284.9 FPS
precision  FP16
platform  Jetson Orin Nano 8GB
jetpack  6.2 (CUDA 12.6)
cloud calls  0 — fully autonomous
# production · computer vision · live IIoT deployment
systemctl status edge-inference
03 — The stack

Every Layer,
Owned

Right edge AI hardware for the brief. Jetson for GPU-accelerated on-device ML. Raspberry Pi and Hailo for cost-sensitive and TinyML deployments. NXP i.MX and Rockchip for IIoT and industrial edge intelligence.

One team across every layer — from custom Linux images and local inference pipelines through cloud services and mobile apps. No integration gaps. No handover risk. Bandwidth-optimised from the ground up.

Hardware
Jetson Orin ★Raspberry Pi 5 ★Hailo-8Coral TPUNXP i.MX 9xRockchip RK3588Custom carrier boards
System
JetPack 6.x ★RPi OS ★Yocto / BuildrootCustom BSPDevice driversReal-time config
Inference
TensorRT ★ONNX RuntimeTFLite / LiteRTHailo SDKeIQ / Neutron NPUCUDARust pipeline
Models
YOLOv8 / v11EfficientDetLlama / MistralWhisperCustom fine-tuneONNX exportINT8 quantisation
Middleware
Conduo (EdgeOps)MQTTgRPCGStreamerLTE-M / NB-IoT
Cloud
AWS IoT CoreLambdaGo servicesOTA pipelineFleet telemetry
Mobile
iOS / SwiftUILive dashboardsPush alertsCoreBluetoothField operator UX
04 — How we work

No Surprises.
Just Delivery.

01
Discovery
Hardware constraints, local inference targets, deployment environment, data privacy requirements. Define what edge intelligence success looks like in measurable terms before anything else.
02
Architecture
System design from silicon to cloud. Hardware selection for your inference budget, on-device ML model strategy, pipeline architecture, integration points — agreed and documented before a line of code is written.
03
Build & Optimise
Iterative delivery with benchmarks at every stage. TensorRT tuning, latency profiling, power budget validation, INT8 quantisation. Real numbers on real edge AI hardware throughout.
04
Handover
Production-ready code, documented architecture, CI/CD pipeline. Optionally retained for ongoing EdgeOps, model refresh cycles, and fleet monitoring across your deployed edge intelligence nodes.
05 — Mission

To build world-class embedded AI and edge intelligence capability in the North — developing local talent, serving Northern businesses, and positioning the region at the forefront of on-device AI and intelligent hardware.

01

Accessibility

Expert embedded AI and edge computing capability without London price tags. Northern businesses deserve access to the same depth of on-device ML expertise.

02

Talent

Keeping Northern engineers in the North. Building on university output and creating pathways into production embedded AI and edge intelligence work.

03

Community

An independent company with a community mission — reinvesting in the Northern tech ecosystem, not extracting from it. Open source everything we build.

04

Scale

Manchester anchor. Leeds, Liverpool, and Sheffield as the network grows. Northern edge AI capability, not Northern limitation.

06 — Who we serve

Edge AI for
Northern Industry

Hardware Startups

Founders with a vision

You have the product concept. You need embedded AI expertise without a full-time hire. Lab access, prototyping support, and fractional CTO capability to get from idea to validated on-device ML hardware.

Scale-ups

Adding edge intelligence

Your product works. Now you need to add on-device AI, connectivity, or real-time inference. We integrate edge computing capability into existing products — without rebuilding from scratch.

IIoT & Manufacturers

Industrial edge AI

Predictive maintenance, computer vision quality inspection, energy monitoring. IIoT and Made Smarter-aligned engagements delivering real operational improvement via embedded AI on real factory floors.

Universities

Spin-outs & research

The bridge from academic prototype to commercial edge AI product. Production hardware experience, real-world autonomous deployment, and a network of Northern tech contacts.

07 — Community pledge

Built
Around
Community.

Community isn't a legal structure. It's how we operate. Every engagement feeds embedded AI and edge computing capability back into the Northern tech ecosystem.

Free Lab Days

Every month, a set number of days are available at no cost to early-stage Northern hardware founders who need access to production edge AI silicon and embedded expertise.

Open Source Everything

Every demo, tool, and case study — from TensorRT benchmarks to on-device ML pipelines — is published on GitHub under manchester-edgeai-lab — free for anyone to use, learn from, and build on.

Monthly Open Sessions

A regular meetup for Northern embedded engineers, edge AI practitioners, hardware founders, and IoT teams. No sales pitch. Just knowledge sharing on on-device ML, local inference, and intelligent hardware.

Knowledge in Public

We publish what we learn — edge AI benchmarks, embedded Linux guides, computer vision tutorials, IIoT case studies. The more we give away, the stronger the Northern edge computing community becomes.

Edge AI — explained

Common
Questions

What is edge AI and how is it different from cloud AI?
Edge AI — also called on-device AI or on-device ML — runs neural network inference directly on local hardware rather than sending data to the cloud. The result is low latency (milliseconds, not seconds), data privacy by design (sensitive data never leaves your site), and autonomous operation that works even without internet connectivity. Cloud AI processes data on remote servers; edge AI processes it where it's generated.
What is local inference and why does it matter?
Local inference means running your AI model on the edge device itself — an NVIDIA Jetson, Raspberry Pi, or dedicated NPU like the Hailo-8 — rather than sending data to a cloud endpoint. Local inference eliminates round-trip latency, reduces bandwidth costs, ensures data privacy, and enables autonomous operation in environments without reliable connectivity.
What is TinyML?
TinyML (Tiny Machine Learning) is the practice of running machine learning models on ultra-low-power microcontrollers and embedded systems — devices with kilobytes, not gigabytes, of memory. TinyML combines model quantisation (INT8, INT4), pruning, and frameworks like TensorFlow Lite to deploy deep learning in the smallest possible footprint. Manchester Edge AI Lab applies TinyML principles across our full platform range.
Which edge AI hardware platforms do you work with?
We deploy across six primary edge computing platforms: NVIDIA Jetson (Orin Nano / NX / AGX) for GPU-accelerated on-device ML; Raspberry Pi 5 and CM4 for cost-optimised deployments; Hailo-8 M.2 for dedicated NPU inference; Coral Edge TPU; NXP i.MX 9x with eIQ NPU; and Rockchip RK3588/RK3568 for industrial IoT. We recommend the right hardware for your inference budget and power envelope.
Can we use AI without sending data to the cloud? Is GDPR compliance possible?
Yes. This is precisely what on-premises AI and local inference are designed for. We deploy open-weight models — Llama, Mistral, Phi, Whisper — directly on your hardware. Data never leaves your network. No third-party cloud processing, no data egress, no GDPR exposure. Our Local AI offering is specifically built for legal, healthcare, and manufacturing environments where data privacy is non-negotiable.
What is IIoT and how does edge AI fit in?
Industrial IoT (IIoT) connects machines, sensors, and systems on the factory floor to enable smarter industrial operations. Edge AI in IIoT moves intelligence from the cloud down to those machines — enabling real-time predictive maintenance, computer vision quality inspection, energy monitoring, and autonomous operation without cloud dependency. We build IIoT edge AI systems aligned with Made Smarter NW and Innovate UK programmes.
Partners & Affiliations
NVIDIA Connect Linux Foundation NXP Partner Made Smarter NW
GO

Build On-Device AI
That Actually Runs.

Most edge AI projects stall between the notebook and the field. This one won't.

Get in touch

Ready to
Ship Something?

Talk to us about lab access, a scoped edge AI engagement, or a technical partnership. We work with Northern startups, manufacturers, and scale-ups who need to turn on-device AI vision into working, deployed product.

Manchester, UK
Contact form
Name