Vennes Data Systems designs data tools for hospitals, biobanks, and clinical research infrastructures across Switzerland. We build software that respects the sovereignty of patient data — by architecture, not by claim.
No cloud transfers. No third-party APIs. Systems deploy entirely within your institutional infrastructure, in full compliance with Swiss data protection law (nLPD).
Validated decisions are immutable. Results remain reproducible across model versions and over time — a property critical to clinical research and regulatory audit.
Explicit versioning, full provenance, transparent operations. Each model output can be reconstructed years later, with its exact computational lineage intact.
A medical entity linker for Swiss biobanks. Built for multilingual clinical reality, deployed on-premise, with full audit trail.
In the late 18th century, French astronomer Charles Messier catalogued the celestial objects that astronomers of his time kept mistaking for comets. His Catalogue des Nébuleuses et des Amas d'Étoiles became the canonical reference — a shared identifier system that let any astronomer, anywhere, point unambiguously to the same object in the sky.
Messier — the system — does the same for the medical text inside biobanks. It transforms free-text disease mentions, written across four languages and a dozen institutional conventions, into a canonical, reproducible, machine-readable identifier in the Mondo Disease Ontology (CC0).
Zuckerkrankheit Typ 2, DT2, E11.9, or type 2 diabetes mellitus — all become MONDO:0005148. One identifier. Four cross-mappings (SNOMED-CT, ICD-10, UMLS, OMIM). Reproducible across institutions, across languages, across time.
Native support for French, German, English, and Spanish. Italian extension planned. Designed for the linguistic fragmentation of Swiss biobank ecosystems.
Existing identifiers from ICD-10, SNOMED-CT, UMLS, Orphanet, OMIM are automatically recognized and resolved via Mondo cross-mappings, bypassing the ML pipeline.
Each prediction is reviewable by a human curator. Validated mappings are cached and immutable — model evolution never overrides curator decisions.
Distributed as a docker-compose stack. Deploys within hospital infrastructure with no external dependency. Zero cloud transfer of patient data.
Validated on independent public benchmarks:
All benchmarks are held-out evaluations on independent public datasets. NCBI Disease is reported as a conservative cold-start benchmark via cross-vocabulary mapping. Methodology and full details available on request.
Messier is currently at TRL 5, transitioning to TRL 6. Core technical components are validated, integrated as a deployable system, and demonstrated in pilot environments. The next phase — operational deployment within a Vaud university hospital — is being prepared in the framework of the SDSC Vaud 3rd Call (September 2026).
Source code under Apache 2.0. Knowledge base built on Mondo Disease Ontology (CC0) enriched with UMLS multilingual aliases.
Hands-on expertise for biomedical data systems. From architecture to implementation, with a bias for sovereignty and reproducibility.
Vennes Data Systems advises hospitals, biobanks, clinical research organizations, and deeptech startups on the design, deployment, and governance of data systems in regulated environments. We bring depth of technical execution and grounded knowledge of the Swiss biomedical ecosystem.
Project-based mandates, recurring retainers for ongoing technical advisory, and focused audits. We work alongside in-house teams, never in their place. Output is always documented, transferable, and useful beyond the engagement.
Current partners include DiData SA (Lausanne), with whom we collaborate on hospital information system integration.
Three principles guide every system we design. They are not values we claim — they are properties our systems demonstrate.
The 2023 revision of the Swiss Federal Act on Data Protection (nLPD), combined with cantonal hospital IT security policies, makes the transmission of patient-linked data to external cloud APIs a non-starter for serious biomedical software. Frontier LLM platforms — however capable — are structurally disqualified from clinical deployment by Swiss law and institutional risk policy.
We treat this as a feature, not a limitation. Our systems are designed to run entirely within institutional infrastructure: a docker-compose stack, a knowledge base on local storage, a model running on a hospital GPU. No outbound API calls. No telemetry. No vendor lock-in. The institution retains full control of its data, its model, and its operational continuity.
Where we rely on external resources, we choose those with public, durable, and license-free terms. Our reference disease ontology is Mondo (CC0). We deliberately avoid SNOMED-CT not because it is technically inferior, but because its licensing creates friction for cross-institutional and international deployment.
Statistical machine learning is inherently probabilistic. But the systems we build are not just statistical engines — they are operational tools used in regulated clinical research, where reproducibility is non-negotiable.
We resolve this tension architecturally. The model evolves; the validations do not. Once a mapping has been validated by an authorized curator, it is cached deterministically and returned identically on every subsequent query — regardless of subsequent model retraining, knowledge base updates, or evolutionary improvements. A study published in 2027 using Messier outputs can be reproduced exactly in 2030.
This separation between an evolving model layer and an immutable validation layer is a core architectural commitment, not an implementation detail.
Every prediction our systems produce carries its full provenance: which model version, which knowledge base version, which alias retrieval ranked the candidates, which curator validated the outcome, when, and against which top-K alternatives. Each deployed system exposes explicit version hashes for all its components.
This makes our systems compatible with the audit requirements of clinical research, regulatory inspection, and reproducibility-grade scientific publication. It also makes them honest about their own evolution: when the model improves and would now predict differently for a historical case, the system surfaces the disagreement for human review rather than silently changing its mind.
The mainstream trajectory of AI in 2026 — cloud-hosted, vendor-controlled, opaque in its inner workings — is incompatible with how regulated industries operate. Healthcare in particular cannot afford systems whose data flows abroad, whose models change without warning, and whose decisions cannot be reconstructed years later.
Vennes Data Systems exists to demonstrate that the alternative is technically practical, commercially viable, and operationally robust. We build systems that institutions can fully understand, fully control, and fully outlive us.
Vennes Data Systems is a Swiss SME founded in 2019, recentered in 2026 on biomedical data infrastructure and applied AI for regulated environments.
The company operates from Vaud and serves clients across Switzerland and, increasingly, the broader BBMRI-ERIC research infrastructure. Our work combines deep technical execution — we write code, we deploy systems, we operate infrastructure — with strategic advisory grounded in lived experience of the Swiss biomedical research ecosystem.
Khalil Roy founded Vennes Data Systems and leads its technical and strategic direction. His prior and current roles include Innovation Officer at Swiss Biobanking Platform (SBP) and consulting engagements across the Swiss biomedical infrastructure ecosystem. He has been involved in BBMRI-ERIC working groups since 2024.
His work spans the design of clinical data systems, applied machine learning for medical text, and architectural strategy for AI in regulated environments.
Vennes Data Systems operates within a network of Swiss biomedical technology partners. We collaborate closely with DiData SA (Lausanne), a software integration specialist for the healthcare sector, on hospital deployment projects. The Messier project benefits from informal advisory by senior figures of the Swiss clinical software ecosystem.
For consulting inquiries, partnership conversations, or technical questions about Messier.
contact@vennes.ai
messier@vennes.ai
linkedin.com/in/khalilroy
Vennes Data Systems Sàrl
Lausanne, Vaud, Switzerland
This site does not use tracking cookies, analytics that collect personal data, or third-party scripts. Your visit is private.