Results from deployments delivered with CoSolve consortium partners — each marked with its status and where the figures come from. More partner case studies are on the way.
SHOWING 8 OF 8
The expert knowledge behind hospital revenue flows, encoded into executable AI workflows that agents run.
Deployed entirely within customer-controlled infrastructure, protecting sensitive patient data.
A health insurer needed extraction, denial classification, summarisation, appeal drafting and policy Q&A over member records that could never leave its private cloud. Evals and data strategy were set before training, then one QLoRA-tuned model shipped that the client owns.
Every release scored against criteria agreed before training.
AI built for Fractal AI's market research and hedge fund operations, supporting its institutional traders.
A physician at one of Europe's leading heart-transplant centres is encoding clinical expertise into executable AI workflows.
Potential impact. Expert-led, evidence-backed workflows — clinical decisions stay with physicians.
Manual compliance checks took 30–45 minutes per loan application. Risk officers now write the rules in plain language; the agent finds the source evidence, marks each item pass, fail or missing, and routes exceptions to a person.
Pilot results. Exceptions always go to a person.
The unit processed around 4,000 child-safety cyber tips a month by hand, each with images, audio and video. Officers write a procedure in plain English and the agent runs it: image origin, regional-language audio, IP tracing to location and ISP, draft ISP requests and the report.
Agentic harness in early deployment. ISP requests are drafted, not sent; officers review the evidence and make every decision.
A small language model that sits inside the phone and operates apps and screens the way a person does — zero tokens, zero cloud, zero data leaving the device. Shown in seven demonstration films, each recorded in one take on one phone.
Measured by The Key 4 AI on a phone, as published at thekey4.ai/proof. Independent benchmarks are in progress.
Questions answered from an organisation's own records, running on its own server. Every answer quotes the source document and shows the page, checked word for word — with scanned files searchable by meaning across languages and scripts.
Demo published by The Key 4 AI at thekey4.ai/proof.
Book a 30-minute briefing with our solution team.