Ubikare, Multiverse Computing and Vicomtech's Q-CLINIC Receives the Ennova Health AI Prize, Announced 2 October 2026: Quantum-Inspired Language Models for Coding Clinical Notes in SNOMED CT and LOINC

Quentir Medicine Monitor

Evidence-based insights for quantum medicine. Published by Quentir Systems LLC · October 5, 2026.

Much of what a hospital knows about its patients sits in free text. A discharge letter, a nursing note or a dictated consultation carries diagnoses, medications and test results in ordinary sentences, and someone has to turn those sentences into structured codes before the information can feed a registry, a quality report or a research query.

Three companies from the Basque Country say they are building a tool for that job. The health software company Ubikare leads Q-CLINIC with the research center Vicomtech and the quantum software company Multiverse Computing. According to a press release distributed through Europa Press on 2 October 2026, the project has received the Ennova Health prize of the Spanish medical and pharmacy titles Diario Médico and Correo Farmacéutico, in the category of artificial intelligence and data management. According to the partners, the system combines quantum-inspired compression of large language models with an extraction pipeline that codes clinical entities using SNOMED CT and LOINC, with HL7 FHIR for data exchange.

The prize was collected by Ángel Díez, chief executive and founder of Ubikare, Gorka Unamuno, head of innovation and strategic alliances at Multiverse Computing, and Gonzalo López, business development manager for digital health and biomedical technologies at Vicomtech. The project is funded by the Basque Government through the Smart Industry Quantum Technologies 2025 program of the regional development agency SPRI. The stated goal is to cut the administrative time of healthcare professionals by up to 40 percent.

For a hospital information officer or a clinical coding manager, the interest lies in two promises: a language model small enough to run on the hospital's own servers, and output that arrives already mapped to the terminologies and exchange formats the hospital uses. Both deserve a close reading, because the public material so far describes an approach and a goal and reports no test results.

What Ubikare, Vicomtech and Multiverse Computing announced on 14 April and 2 October 2026

The partners first presented Q-CLINIC on 14 April 2026, World Quantum Day. The launch announcement, republished by the Spanish communication directors' association Dircom on 15 April, describes the problem in plain terms: valuable medical information is produced as free text or voice, and structuring it by hand creates administrative load and blocks data-driven medicine.

The announcement lists three properties. The first is efficiency: the partners apply what they call quantum LLM compression, citing techniques such as variational quantum circuits and quantum-inspired tensors, to reduce the computational cost and energy use of the language models. The second is privacy: the compressed models are meant to run on-premise on conventional hardware, so clinical data are processed without leaving the healthcare environment. The third is accuracy: the pipeline built with Vicomtech is meant to identify clinical entities and code them automatically according to international standards, which the announcement lists as SNOMED CT, LOINC and HL7 FHIR.

Q-CLINIC extends Ubikare's existing platform NAIHA, short for Natural and Artificial Intelligence Health Assistant. Ubikare's website describes NAIHA as software classified as a Class IIa medical device under EU health regulation. The two announcements do not say whether the Q-CLINIC components fall under that classification, which hospital sites have used the system, or which datasets it has been tested on. The 2 October release repeats the April description and adds the award; it reports no accuracy, coding agreement or time-saving figure measured in a hospital.

Quantum pillar: not applicable. Technology readiness: not applicable. The partners have published no test results, accuracy figures or clinical sites for Q-CLINIC, so the cited public material does not establish a readiness level; what is public is a stated target of 40 percent less administrative time, and the compressed models run on ordinary computers with no quantum processor involved.

How quantum-inspired tensor-network compression shrinks a language model to run on hospital servers

The word quantum in Q-CLINIC refers to a family of mathematical tools taken from quantum physics. Multiverse Computing's compression method, called CompactifAI, uses tensor networks, a way of factoring very large arrays of numbers that physicists developed to describe quantum many-body systems. Applied to a language model, the method replaces the large weight matrices of selected layers with chains of smaller tensors that keep the most important correlations and discard the rest. The calculation runs on ordinary processors.

The method is public. In the CompactifAI preprint by Andrei Tomut, Saeed S. Jahromi, Román Orús and colleagues, first posted to arXiv on 25 January 2024, revised on 13 May 2024 and published in the proceedings of the ESANN 2025 symposium in Bruges, the authors report that combining tensor-network compression with quantization reduced the memory size of a 7-billion-parameter LlaMA model by 93 percent and its parameter count by 70 percent, with training 50 percent faster, inference 25 percent faster and an accuracy drop of 2 to 3 percent. Those numbers come from general-language benchmarks. They say nothing yet about how a compressed model handles Spanish or Basque clinical shorthand, drug names, negations such as "no signs of pneumonia", or the abbreviations that vary from ward to ward.

For a hospital, the practical appeal is local deployment. A model that fits on a server inside the hospital network keeps patient text away from external cloud services, which can make the data-protection assessment under the GDPR simpler and may lower running costs; neither effect has been shown for this project. The trade-off is that every percentage point of accuracy lost in compression can turn into a coding error that a human must catch later. A buyer will want to see the clinical accuracy of the compressed model next to the uncompressed one, on the same notes.

This Monitor has followed Basque quantum work in medicine before. In August it read the Tecnun study that generated synthetic myelodysplastic-syndrome patients on IBM's ibm_basquecountry processor, which ran on real quantum hardware. Q-CLINIC sits at the opposite end of the spectrum: its quantum element is mathematical inspiration, and its value will be measured like any other clinical language tool.

Which coding standards Q-CLINIC targets: SNOMED CT, LOINC and HL7 FHIR

The three standards in the announcement do different work. SNOMED CT, maintained by SNOMED International, is the clinical terminology that gives diagnoses, findings and procedures a precise code. LOINC codes laboratory tests and clinical observations. HL7 FHIR is the exchange format that moves structured data between systems, so a coded diagnosis can travel from a note into an electronic health record, a registry or a research platform.

Automatic coding into these standards is an established and difficult task. Errors usually come from context: a diagnosis that is only suspected, a condition that belongs to a family member, a finding that has been ruled out. A system that reduces administrative time is only useful if its codes are correct often enough that clinicians and coders review them quickly and trust them.

What a hospital buyer should ask Ubikare before a Q-CLINIC pilot

The award shows that a Spanish medical trade jury found the project promising. For a procurement decision, four questions matter more. First, what is the coding accuracy on the hospital's own notes, measured against trained human coders, and how much does it change between the compressed and full-size model? Second, does the 40 percent time saving come from a measured pilot or remain a target? Third, does Ubikare's stated Class IIa classification of NAIHA cover the Q-CLINIC extraction function, or will it need its own conformity assessment? Fourth, which languages and specialties has the pipeline been tested on?

The next evidence worth watching for is a published evaluation on real clinical notes from a named hospital, with agreement figures for SNOMED CT and LOINC codes and a measured change in documentation time. Until then, Q-CLINIC is a funded and award-winning project with a credible compression method behind it and its clinical performance still to be shown.

Sources

Primary source: press release of Ubikare, Multiverse Computing and Vicomtech on the Ennova Health prize, distributed through Europa Press, Madrid, 2 October 2026. Also drawn on: the Q-CLINIC launch announcement of 14 April 2026 as republished by Dircom, Ubikare's description of NAIHA, the CompactifAI paper of Tomut, Jahromi, Orús and colleagues (arXiv 2401.14109, ESANN 2025), and the public pages of SNOMED International and HL7 FHIR. Readiness and implications are the Monitor's editorial assessments.

1. a press release distributed through Europa Press on 2 October 2026
2. The launch announcement, republished by the Spanish communication directors' association Dircom on 15 April
3. Ubikare's website
4. the CompactifAI preprint by Andrei Tomut, Saeed S. Jahromi, Román Orús and colleagues
5. SNOMED CT, maintained by SNOMED International
6. HL7 FHIR

The Quantum Medicine Evidence Register

One briefing reads a single study. The Evidence Register keeps the whole field in view for you.

It is a living evidence-intelligence system for quantum medicine: claims tracked across regions and languages, each placed on the evidence ladder from theory to simulation to hardware experiment to real-world use, with peer-reviewed work, preprints, and company announcements weighted differently and their sources always at your elbow. Where a post like this one captures a moment, the Register keeps the picture current, updated continuously and synthesized weekly, so you can walk into a board meeting or a diligence call knowing exactly what the record supports today and where it stops.

See what the Evidence Register covers →