We build the companies we want to exist.
Creative Pandas is an independent Swiss holding company. We start, fund and operate AI-native software ventures — clinical data infrastructure, privacy-first assistants, company intelligence, quantitative research — and we build them ourselves rather than selling advice about them.
The ventures
Each venture is operated independently, with its own product, customers and roadmap. What they share is the engineering bar and the infrastructure underneath.
MedBackend
Multi-tenant healthcare infrastructure built on FHIR from the ground up, so clinical software can exchange patient data without every integration becoming a bespoke project. In production, with a deployment running in Estonia.
Schatzi
A conversational AI assistant hosted entirely in Switzerland, for people and companies who need to know where their conversations are processed and stored.
AltSignals
A managed data product that turns public professional-network activity into a structured company feed for sales and research teams, with a self-service trial portal and a documented API.
OfficeLabs
A LibreOffice fork where AI lives inside the document engine rather than bolted on as a plugin — so assistance works on the real document model, offline and under your own control.
VaxMe
Photograph a paper vaccination booklet and get structured, FHIR-shaped records back. Vision-language models handle the handwriting that OCR alone cannot.
Lindest
Quantitative research into systematic trading signals across European and US small-cap equities, built on reproducible data pipelines rather than discretionary calls.
Projects we contribute to but do not own.
FastSME
An open-source portfolio bringing enterprise software patterns — CRM, ERP, planning, analytics — to companies normally priced out of them. Patrick contributes to the project; it is independent of Creative Pandas.
How the holding works
Swiss by construction
Incorporated in Switzerland and hosted here. For the ventures handling clinical or personal data, where the data is processed is a product requirement, not a compliance checkbox added afterwards.
Operators, not advisors
We hold equity in what we build and run it in production. Every venture in the portfolio has real users or real customers — none of them are prototypes kept alive for a pitch deck.
Shared foundations
Ventures reuse the same deployment, authentication, secrets and model-routing stack. A hard problem solved once in healthcare data becomes free for the next venture that needs it.
Open where it helps
Infrastructure that is more useful shared than owned gets released publicly, and we contribute to open-source projects beyond our own portfolio.
Engineering practice
The work across the portfolio keeps landing on the same set of problems. These are the ones we have solved repeatedly and can build on immediately.
Clinical data interoperability
FHIR modelling, multi-tenant isolation, and moving records between systems that were never designed to talk to each other.
Language models in production
Model routing, Swiss-hosted inference, prompt and evaluation pipelines — including vision-language models reading documents that defeat plain OCR.
Autonomous agents at scale
Long-running agent fleets with supervision, retry semantics, and the monitoring needed to know an agent is degrading before a customer does.
Data pipelines that survive contact
Collection, parsing and enrichment pipelines that run continuously against sources that change without warning.
Published work
Infrastructure we found more useful shared than kept. Released under the Creative Pandas organisation on GitHub.
zefix-mcp
An MCP server for Zefix, the Swiss Federal Commercial Register. Look up any Swiss company by UID, name, or canton — including legal form, registry office and SHAB publications — from inside an AI assistant.
More of our work opens up over time — follow the organisation on GitHub.
Building something adjacent?
We talk to founders, operators and prospective partners — particularly where a venture in the portfolio already touches the problem.