Drafting discharge letters has historically been one of the heaviest and least rewarding bureaucratic hurdles for hospital medical staff.
Every time a patient completes an inpatient stay, doctors must retrieve and summarize a vast amount of clinical information from consultations, clinical courses, tests, and pharmacological therapies, converting them into a formal document intended for the primary care physician or specialist. It is a long, repetitive process with a high risk of cognitive overload, which takes valuable time away from direct patient bedside care. To address this challenge, a web app named Cyrano was created, designed to introduce advanced artificial intelligence models into the daily hospital workflow, making the drafting of clinical documents fast, accurate, and completely secure.
A Synergy Supported by Innosuisse
Cyrano was born out of a strategic alliance combining software development, academic research, and the clinical practice of a large hospital authority. The project was developed by Ex Machina in close collaboration with the Ente Ospedaliero Cantonale (EOC) and IDSIA USI-SUPSI (Dalle Molle Institute for Artificial Intelligence Studies); demonstrating the strategic value of the initiative is the exclusively financial support from Innosuisse, the Swiss Federal Innovation Agency, aimed at supporting the research side of the project conducted by EOC and IDSIA USI-SUPSI.
In this integrated research and development journey lasting roughly two years, each partner provided a fundamental contribution. While Ex Machina designed the software architecture and orchestration platform, and EOC provided the testing ground and field clinical expertise, the academic contribution of IDSIA USI-SUPSI was key in defining extremely rigorous scientific and methodological standards. The primary goal was to ensure that every discharge letter was generated with maximum clinical accuracy and consistency, while simultaneously developing innovative systems capable of providing suggestions and corrections that the physician can accept or reject.
Technological Architecture, Multilingualism, and “LLM-as-a-Judge” Evaluation
The integration of Cyrano into ward processes occurs completely transparently through connection with the hospital’s Single Sign-On system and interfacing with the Electronic Health Record. The patient’s health data is extracted and reorganized into a structured canonical format, specifically designed so that the language model can process it without ambiguity. The platform demonstrates extraordinary operational flexibility by managing customized generation rules for individual medical specialties: in the pilot departments of Internal Medicine and Surgery, for example, differentiated standards and paragraphs were defined that reflect the real management needs of the different clinical teams. Furthermore, to perfectly address the complexity of multilingual healthcare contexts such as Switzerland and international environments, the web app is natively prepared and already fully operational in multilingual mode, currently supporting Italian, French, and German.
One of the most innovative and elegant elements of Cyrano’s architecture lies in its automated self-checking mechanism based on the technical LLM-as-a-judge pattern. Once the first draft of the discharge letter is generated, the system launches a second independent session driven by a twin language model, with context and instructions dedicated to validation. This system scrutinizes the generated text by comparing it directly with the original clinical data in the patient’s record. This cross-analysis enables the timely detection not only of writing inaccuracies, but also of potential inconsistencies present at the source in the hospital documentation, such as previously entered conflicting clinical notes. By flagging these anomalies to the doctor as targeted suggestions, Cyrano not only speeds up the drafting of the document’s prose, but concretely assists healthcare staff in cleaning up the clinical database directly at the source.
Data Privacy and the Protection of Sensitive Health Data Form the Bedrock of the Entire Infrastructure. Cyrano is installed entirely on-premise, confined within the hospital authority’s private, protected network. The architecture developed by Ex Machina is designed to ensure maximum independence from individual proprietary cloud vendors, utilizing open-weight models sized to the hardware infrastructure available to the hospital. An intelligent orchestrator manages request concurrency to prevent saturation and allows performance benchmarks to be periodically repeated as new and evolved versions of language models emerge.
Impact on the Wards: The Clinical Trial with the EOC
The transition from theoretical model to real clinical practice saw the Ente Ospedaliero Cantonale act as a pioneer. The first phase of the trial was conducted under the scientific supervision of three senior chief physicians in the Internal Medicine and Surgery departments, who rigorously evaluated the AI-generated output based on a precise, agreed-upon metric rubric.
The results from the clinical evaluation proved to be extremely positive, demonstrating that Cyrano offers extraordinary efficacy for low and medium-complexity stays, which represent roughly 80% of all hospital caseloads. In light of the initial excellent feedback, the testing phase was recently extended to a group of 13-15 resident physicians who use the web app daily in ward operations to precisely quantify time savings. The feedback gathered among doctors remains enthusiastic: automating a task perceived as routine and secretarial finally returns valuable time to healthcare professionals to dedicate to human contact and medical care.
From Research to Market: Scalability and Future Prospects
The true strength of the platform developed by Ex Machina lies in its generic and highly scalable nature. Not being tied to any specific infrastructure, the web app can be easily adopted by any hospital facility, both public and private. The commercialization model is based on a transparent pay-per-use formula, proportional to the actual number of discharge letters generated, accompanied by customization and IT integration packages. Thanks to the combination of scientific rigor, technological security, and real clinical utility, Cyrano stands as a candidate to become a benchmark for digitalization and efficiency in hospital healthcare.
