Edinburgh is the UK's second financial centre by a distance, weighted towards asset management and life assurance rather than the trading floors of London. Add the banks, the university's informatics base, and an energy sector shared with Aberdeen, and the mix produces work with an unusual character: heavy on documents, heavy on evidence, and staffed by people who will ask what the evaluation set looks like.
Fund managers and insurers arrive with reporting. Quarterly client reports, fund factsheets, suitability files, regulatory returns — each assembled by a person from numbers that already exist somewhere else, and each carrying a liability if it is wrong. Drafting is the easy part; the value is in the checks around it. The pattern is described in regulatory reporting, and the surrounding control structure in our risk and compliance architecture.
The research connection shows up as a specific request: prove it. Teams here are comfortable with what a language model can do and sceptical about whether a particular system does it well enough to trust, which we regard as the correct posture. Almost every Edinburgh engagement starts with an evaluation harness rather than a demo — see evaluation — because a number everyone agrees on ends the argument that a demonstration only postpones.
Energy work, whether from Edinburgh offices or Aberdeen operations, tends towards asset records and maintenance history rather than language: decades of inspection reports about a small number of very expensive things. That is closer in shape to predictive maintenance than to anything in financial services.
