UKB spin-off · Digital health
A picture is worth more than a thousand words.
Clear Presentation always wins over simple Extraction.
Redoca is an agentic harness that interfaces neatly with a KIS's extraction pipeline and local LLMs to present clinicians with visual answers to their questions.
The problem
What happens if the Chart does not answer the question?
A clinical dashboard ships with a fixed menu of views, whatever a vendor thought to build in advance. The question in front of a clinician right now is often not one of them. And the more options a dashboard has, the more difficult its UI is to navigate.
In practice, that means reading the numbers off by hand. But we believe people are visual animals, built to see a pattern, not scan a table.
Redoca skips the menu. Ask in plain language, and the view is built for that question, verified before it renders. It runs on‑premise, so nothing has to leave the hospital to make that possible.
The method
Four checks stand between a note and a chart.
Built on the KIS's own extraction, not around it.
Redoca's harness sits on top of the hospital's existing extraction pipeline and local, open-weight LLMs. Each field arrives with its unit, date, source document and exact character span.
A German clinical terminology layer.
Matched against existing KIS vocabulary layers or open terminology sets, depending on the hospital's setup.
The agent adapts a template. It doesn't improvise one.
One of hundreds of verified, parameterised base views is selected and adapted, rather than generating a chart from scratch.
Nothing renders until it's checked.
Schema, unit, range, temporal order and provenance are all verified first. A failure blocks the render. It doesn't get shown anyway.
Deployment posture
Redoca runs on-premise, inside the hospital's own infrastructure. Patient text never reaches a cloud model, and nothing is written back to the record.
How it works
One exchange, four beats.
Ask
A clinician types, or dictates, what they want to see, in plain language.
Verify
Schema, unit, range, temporal order and provenance are checked before anything renders.
Inspect
A dashboard appears, showing patient against cohort with medication changes marked. The agent asks back if anything's ambiguous.
Trace
Click any point. The exact sentence it came from lights up in the source note.
At the bedside
The same exchange, in a clinician's pocket.
Most of a ward round happens away from a desk. Redoca renders the same way on a phone at the bedside as it does on a monitor. No separate mobile view to maintain, nothing lost in translation.
The team
Technical, medical, legal. All inside the hospital.
A mixed team with real documentation workflows on hand, and a regulatory path considered from day one, with BoHAIMe support and access to a certified ISO 13485 quality management system.
Paul Johannssen, MSc
CEO & CTO
PhD candidate in software development / medical informatics, UKB.
PD Dr. Theodor Rüber
Chief Medical Officer
Medical research group leader at UKB.
Dr. jur. Amelie Böcker
Chief Legal Officer
Scientific lead at ZMDT Bonn.
Where this is headed
From concept to the ward.
- Today
Concept validation
Early research stage. No MVP yet: three templates, one clinical indication, real German discharge letters, local inference, and a working click-through to source.
- 2026 – 2027
Functional prototype
Semantic integration with German diagnosis and procedure catalogues (ICD-10-GM, OPS). Pilot at UKB Bonn in epilepsy.
- 2027 – 2029
Regulatory path
MDR conformity assessment, then market entry, with a certified ISO 13485 quality management system already in place.
Get in touch
Building, piloting, or funding? Let's talk.
Redoca is looking for co-founding software developers with exposure to AI, and for hospitals interested in an early pilot.
paul.johannssen@ukbonn.de