Skip to content

Work

Innovation projects we have delivered for our customers, from proofs of concept and investor demos to production systems.

  1. Natural disaster protection through a smart contract insurance platformA decentralised platform that replaces slow, intermediary-heavy risk transfer with smart contracts, tokenisation and a global risk pool, built for natural disaster insurance and live in production.
  2. A project and quality management platform for regulated pharmaA modular project and quality management platform for pharma: we helped strategise, pilot and scale a new product to speed up IT project delivery, and developed it fully, including for the client’s investor pitches.
  3. An e-games tournament and competition platformA scalable online platform for e-sports competition: players of similar level are matched, communities connect and compete, and the option to play for stakes adds to the incentive. A working product was delivered to the customer.
  4. A helm for a fully electric boatA first working prototype of a complete helm for a fully electric boat: navigation, charts and live analytics, wired into the vessel’s own hardware and demonstrated on the water in Spain ahead of the client’s funding round.
  5. A standardised trade recap application for a commodity trading houseOne application that creates every trade recap in the same standard form, delivered for a commodity and energy trading company in the UAE.
  6. Business process management across the trade lifecycleThe whole trade lifecycle documented and standardised, with KPIs and controls that raise process maturity and support growth into new markets.
  7. A bunkering blotter integrated with AspectA blotter for bunkering trades, connected to Aspect, so deals are captured once and move faster from the desk to the books.
  8. A trade KPI report across the trade lifecycleOne report that tracks enterprise KPIs across every stage of the trade lifecycle, from deal capture to settlement.
  9. AI avatar training across the trade lifecycleImmersive training led by AI avatars for the people who touch a trade: traders, operators, contract administration, risk, finance, legal, tax and others.
  10. A multi-channel AI concierge, in productionA live AI concierge answering customers on five channels at once, with model routing that fails over on cost and reliability and a dashboard that shows staff what each channel costs.
  11. Agent orchestration for the back officeFive role-bounded agents process accounts-payable invoices end to end, with fraud and duplicate checks built in and a human operator deciding every case the system is unsure about.
  12. An adversarial evaluation harness for allergen safetyA 117-case adversarial test set for allergen questions, wired into CI so no change to the model, prompt or retrieval ships unless every case still passes.
  13. A grounded knowledge copilot for field serviceA retrieval system for technicians that cites its sources or abstains, scores each answer, recommends resolve or escalate, and drafts knowledge-base updates that a person approves before anything is published.
  14. A cold-chain command centreA command centre for a temperature-controlled distribution network: an operator asks what is happening in plain language and sees excursions, delays and stockout risk on a map.
  15. Visual reconciliation for returned kitsA vision system that reconciles a returned kit against its manifest item by item, with a bounding box, confidence score and condition class per item, and a replenishment recommendation attached.
  16. A tiered document intelligence pipelineA contract-intelligence pipeline that spends its strongest model only where the task needs it, with per-tenant budget caps built into the architecture.
  17. Anonymise before inferenceA pipeline that replaces every real name, address and identifier with a stable alias before a model sees the content, checks independently that nothing leaked, and reverses the substitution only locally.
  18. On-device inference, with nothing leaving the deviceSpeech and language models that run entirely on the device, with no network call at inference time, for settings where connectivity is unreliable and the data should not leave the hardware.