R&D
Including the ones that did not go anywhere.
R&D that only ever reports successes is marketing. These are the prototypes and internal tools we have built, with what actually happened to each of them.
Fraud response copilot
A working response to a regulator-mandated 30-minute fraud handling window: triage, evidence assembly and case narrative drafted from transaction data, built end to end on synthetic data.
Proved the window is achievable with assisted triage. Built as a white-labellable demo, not yet deployed with a bank.
Election intelligence platform
Five modules over a national polling-unit dataset of around 176,000 units, covering result intake, anomaly detection and field reporting at national scale.
Prototype and technical proposal completed. Not taken forward commercially.
Natural-language Office automation
A declarative command layer letting people describe a document or deck change in plain language instead of hunting through ribbons, over 50-odd composable commands.
Reviewed in early 2026 against a market where Copilot now ships inside Office. We parked it rather than compete head-on, and kept the parts that modernise legacy VBA and VSTO estates.
Content engine
Authoring, magic-link client review and versioned publishing for the content that goes into client sites, so review cycles stop happening over email attachments.
In use on live client engagements.
Get in touch
Got something that needs proving before it needs building?
Feasibility work, a working prototype against your real data, or a straight answer on whether the idea holds up.