Bay Information Systems is an independent AI consultancy based in Cardiff. It was incorporated in 2013 out of machine learning research at university, and its early work applied machine learning to embedded systems and microcontrollers, where memory and power budgets leave little room for waste. In 2019 the company moved from its own product development into consulting.
The work now has three parts. We advise organisations deciding whether and how to use AI on data they control: what to build, how to measure it, and where it should run. We publish the reasoning behind that advice in the article library, which covers private inference, retrieval, evaluation and the cost of running models. And we maintain open-source tools from the same work: Marigold, a self-hosted inference API for open-weight models, and the runfox, json-logic-path and dynawrap libraries.
Most engagements start before anything is built, with the questions that decide whether an investment will work: what the data contains, whether the problem is defined well enough to measure, and what a system would need. The result is a written document the organisation can act on and share internally. (The usual starting point is a one-week AI roadmap; the free readiness assessment covers the same ground in less depth.)
Specialists join engagements where the work needs them (engineers, designers and domain experts), and the principal remains the point of contact throughout. Building products alongside consulting keeps the company close to the reasons systems fail after they work technically: adoption, pricing, and the commercial pressure that gets projects cut.
AI strategy and engineering consultant
Edward has built and deployed machine learning systems for two decades, across academic research, regulated industry and early-stage products. Clients have included the Bank of England, the Office for National Statistics and Admiral plc, and startups in health, adtech and life sciences. Recent work covers retrieval-augmented generation, knowledge graph construction and private cloud deployment on AWS, and he has presented benchmark work on open-weight models at AI Wales.
Selected publications
(To discuss a project, use the contact form or email ed@bayis.co.uk. Engineers and strategists interested in associate work can read work with us.)