Most organisations building AI have two groups involved with data and a gap between them. Data engineering owns the pipelines, warehouses, and infrastructure. The AI or data science team owns models. Between the two sits a set of responsibilities that neither fully owns: deciding what data a model needs, getting it into a usable state, […]

Most organisations sitting on years of proprietary data eventually ask the same question: can we use it to make a model that understands our business? The answer is often yes, and the path there is more constrained than the framing suggests. Fine-tuning is frequently the wrong tool for what teams actually want, and where it […]

Computer vision arrived in factories and fields later than in consumer applications, and it is now spreading quickly: defect detection on production lines, equipment condition monitoring, crop health assessment, livestock monitoring, and yield estimation. These deployments share a set of data problems that differ meaningfully from the ones consumer computer vision solved, and teams that […]

Geospatial AI has moved well beyond mapping. Satellite and aerial imagery now feeds models that detect infrastructure change, monitor environmental conditions, assess damage after events, track development, and support intelligence workflows. These models learn from annotated imagery, and the annotation is unlike most other computer vision work: the objects are small, the imagery is enormous, […]

Financial crime compliance has a structural problem that automation has changed but not solved. Transaction monitoring systems generate enormous alert volumes, the large majority of which are false positives, and every alert has to be reviewed by someone because the consequences of missing a real one are regulatory and severe. Adding AI to this has […]

Every platform that hosts user content or deploys generative AI eventually builds a trust and safety function, usually later than it should have. The trigger is typically an incident: something harmful published, a regulator's letter, or press coverage. What gets built in a hurry after an incident tends to be a moderation queue, which addresses […]