Internal product research

Ideas under investigation.

Our internal work explores how AI might support evidence-intensive tasks. Each concept has a stated status and a clear boundary around its claims.

Concept interface connecting sample business data to a prototype and review queue — Sample, prototype, review · AI-generated concept illustration

Internal exploratory work

Sentinel

Illustrative laboratory equipment beside a procedure and incident-review workstation — Laboratory incident review · AI-generated setting, not an actual facility

Sentinel addresses laboratory incident support: making the relevant procedure and incident context faster to examine together. The research direction concerns source visibility, missing information and the boundary between assistance and accountable human judgment.

It is not presented as a deployed or validated product. Performance, workflow suitability and any operational requirements would need to be established through a separately scoped investigation.

Explore Sentinel

Reference architecture

Private RAG

Private RAG is reference work around retrieval-augmented generation (RAG): finding relevant passages from a defined collection and supplying them as context for a model's answer. The architecture offers a way to discuss where source selection, permissions and answer generation need to be examined.

Retrieving relevant information does not guarantee that an answer is correct or supported. A useful experiment would separately assess whether the right evidence was found and whether the resulting answer used it accurately. The reference architecture itself is not evidence of deployment, privacy assurance or benchmark performance.

A diagram separating answer claims with source support from claims without it — Check the support for each claim · Concept diagram

Proposed research concept

EvidenceDesk

A proposed concept for inspecting the relationship between claims and supporting material. The question it tests: whether reviewers identify unsupported statements, inspect source passages and recognise uncertainty more reliably with it than without.

EvidenceDesk is not deployed or benchmarked. Interface ideas and proposed assessment criteria should be understood as research intentions, not demonstrated improvements in review quality.

Archived proof of concept

ClinicFlow

ClinicFlow is retained as an archived proof of concept. It is not an active clinical product offering and carries no claim of clinical integration, validated performance or operational deployment.

The archive distinguishes historical exploration from the studio's current focus on limited research, experimental prototypes and model evaluation.

A future research direction

Questions beyond our current AI concepts.

We are also exploring a possible direction in quantum computing research. The emphasis is on problem selection, classical benchmarks and a credible feasibility study, with no quantum product or validated result being presented.

Read the quantum computing exploration ↗