Computer Vision
Outcomes
Inspection at a rate and consistency people cannot sustain
Every item checked by identical criteria, continuously, with a record of the judgement.
Detection of things that are easy to miss
Small defects, subtle changes over time, events in footage nobody has time to watch.
Documents processed without keying
Invoices, forms, identity documents and handwritten records turned into structured data.
An honest read on your imaging conditions
Frequently the highest-value output of early work is that your cameras, lighting or angles need to change before any model will perform. Better to hear that in week two.
What we build
Defect and quality inspection for manufacturing lines, tuned to your tolerance for false positives versus missed defects — a business tradeoff we make explicit rather than optimising for a single accuracy number.
Object detection and counting for inventory, footfall, vehicles and assets.
Document understanding — OCR, layout analysis, table extraction, handwriting — including the validation logic that catches misreads before they enter your systems.
Video analytics for safety compliance, process monitoring and event detection, with the privacy controls that use case demands.
Condition and change monitoring across time-series imagery, for infrastructure, agriculture and site progress.
Annotation pipeline and dataset management. Vision projects are dataset projects. We build the labelling workflow, quality controls and versioning, because model architecture matters far less than data quality.
How it works
Week 1 — Imaging assessment. We examine your actual images or footage under real conditions. Lighting, resolution, angle, occlusion, variation. This determines feasibility more than anything else, and it is where we most often recommend fixing the capture setup first.
Weeks 2–4 — Dataset construction. Collection, annotation, quality review, and building a held-out set that includes the hard cases. Rare-event problems need deliberate collection of the rare event, which frequently means waiting or staging.
Weeks 4–6 — Modelling. Starting with pre-trained architectures fine-tuned on your data. Training from scratch is rarely justified. Evaluated against realistic conditions, not a clean subset.
Weeks 6–8 — Edge or cloud deployment. Model optimisation, quantisation where running on-device, integration with your line, camera system or workflow.
Weeks 8–10 — Field validation. Real conditions, real operators, failure collection. Vision systems consistently encounter situations absent from the training set, and this phase is where they surface.
Ongoing. Monitoring, periodic retraining as conditions and products change.
Technology
Models: YOLO family for detection, segmentation architectures where boundaries matter, vision transformers where the problem justifies them, and increasingly multimodal models for document and scene understanding — which have made a category of previously custom work straightforward.
OCR and documents: layout-aware extraction combining traditional OCR with vision-language models, validated against structural rules.
Deployment: NVIDIA Jetson or comparable edge hardware where latency or connectivity requires local inference; cloud inference where it does not. Quantisation and pruning to fit the hardware you actually have.
Annotation: CVAT, Label Studio or similar, with review workflow and inter-annotator agreement checks.
Where this applies
Strongest where the visual task is repetitive, high volume, and currently done by people whose attention degrades over a shift.
Weakest where conditions vary wildly, examples of the target condition are extremely scarce, or the required accuracy leaves no room for review.
How we scope and price
Fixed scope, quoted after an imaging assessment. Cost is driven overwhelmingly by dataset work — whether usable annotated data exists, and how rare the target condition is. Edge deployment adds hardware and optimisation effort. We assess before quoting because vision projects vary more in cost than any other category we work in.
Frequently asked questions
For fine-tuning a pre-trained model on a well-defined task, often several hundred to a few thousand per class — but the number matters less than coverage of real variation. A thousand images from one lighting condition is worse than three hundred across all of them.
Often, and the assessment tells you. Where it cannot, the fix is usually cheaper than people expect — better lighting or a changed angle frequently outperforms a better model.
No. Edge deployment is standard where latency, bandwidth or privacy requires it, and modern edge hardware handles most inspection workloads comfortably.
Determined by your imaging conditions and the difficulty of the distinction, so we assess before quoting a figure. We also frame it as the tradeoff it really is: catching more defects means more false alarms, and where you sit on that curve is your decision.
By design: on-device processing where possible, face and identity blurring, retention limits, and processing only what the use case requires. In many jurisdictions this is a legal requirement rather than good practice, and it shapes the architecture from the start.
Performance degrades and monitoring flags it. Retraining on new examples is routine and planned for, not an emergency.
More AI services
AI Strategy Consulting
Turn scattered AI ambition into a sequenced, costed plan. We decide what to build, what to buy, what to ignore, and in what order.
AI Readiness Audit
A 3–4 week assessment of your data, systems and processes that returns a ranked, costed list of AI use cases and an honest verdict on what you can deploy now.
Agentic AI Automation
We build AI agents that complete multi-step work inside your systems — with defined scope, human checkpoints, and evaluation. Deployed to production, not demos.