A German hospital pilots an on-premise Mistral Small 3.2 system from myScribe that drafts discharge summaries from FHIR patient records for clinician review
An ongoing pilot in a German hospital runs Mistral Small 3.2 fully on-premise under GDPR constraints to draft discharge summaries from structured FHIR patient data. Clinicians edit and verify every draft before it enters official documentation. The EACL 2026 industry-track paper compares four input encodings on 31 patients from the Internal Medicine department, with clinician ratings; JSON encoding performed best.
Overview
An ongoing pilot in a German hospital runs Mistral Small 3.2 fully on-premise under GDPR constraints to draft discharge summaries from structured FHIR patient data. Clinicians edit and verify every draft before it enters official documentation. The EACL 2026 industry-track paper compares four input encodings on 31 patients from the Internal Medicine department, with clinician ratings; JSON encoding performed best.
This entry has 13 published fields tied to exact passages in an immutable source capture.
Inspect the highlighted sourceThe challenge
At the end of hospital treatment the many electronic health records created for a patient must be summarised into a discharge summary, which the source calls a time-intensive bottleneck for doctors.
The solution
A zero-shot system feeds structured FHIR patient data, encoded as text, to an out-of-the-box Mistral Small 3.2 running fully on-premise in the hospital to draft German-language discharge summaries. Clinicians edit and verify every draft before it enters official documentation.
Reported business value
The source frames this as an ongoing pilot and reports no cost or time savings. In an exploratory evaluation on 31 patients, JSON input encoding gave the strongest overall clinician-rated summary quality; the clinicians also found the system often failed to formulate correct follow-up steps, producing hallucinated text.
Sources
Open any source and check the claim yourself — that is the point of the register.
This record was researched and written with AI assistance, and its claims were checked against the sources above. (EU AI Act art. 50 transparency notice.)
Other healthcare entries in the register.
Less paperwork, more time for people: Azure OpenAI at Landsberg Hospital
Klinikum Landsberg am Lech, a German acute-care hospital, deployed an AI language solution built on Azure OpenAI with partner Pexon Consulting that transcribes doctor-patient consultations in real time via tablet, recognizes medical terms, structures the content, and transfers it directly into the hospital information system.
Novo Nordisk builds AI drug discovery platform on Azure with Microsoft Research
Novo Nordisk partnered with Microsoft Research to build an AI platform on Azure AI and data stacks spanning regulatory affairs, early research, drug discovery and trial design, using Azure OpenAI Service, Azure Cosmos DB and Azure Kubernetes Service, with Power BI and Power Apps for collaboration. The platform includes a copilot for researchers, shared reasoning-chain templates, and governance/auditing of how data and models are used. The teams published early results on predictive AI models for cardiovascular disease risk detection, including an algorithm that Novo Nordisk says predicts patients' cardiovascular risk better than the best clinical standards, drawing on more than 100 years of insulin research data.
BAYADA Builds a Unified, AI-Ready Platform for Compassionate Care
BAYADA Home Health Care is consolidating three separate legacy data platforms (multiple practice management systems, an on-prem ODS, and Snowflake) and 65+ enterprise data sources into a single Databricks Lakehouse under its Data Modernization program, using medallion tiers, Lakeflow Jobs, Asset Bundles, and Unity Catalog governance. As part of the migration, BAYADA uses an LLM-powered code converter and Databricks Assistant to automate SQL and stored-procedure translation, and applies a machine learning-based data mastering accelerator to create golden Client, Payor, Candidate, and Referrer records. BAYADA reports roughly 30% better workload performance and cost efficiency versus its prior Snowflake environment, and is laying a governed foundation for AI agents supporting payroll validation, compliance monitoring, and operational insights, separately from its roadmap for AI-assisted chart review and risk prediction.
Heidi Health fine-tunes NVIDIA Parakeet/Nemotron ASR, cutting clinical transcription latency 75% and costs 64%
Heidi Health, which processes 2.4 million clinical consultations weekly across 110 languages, fine-tuned the open NVIDIA Parakeet V2 speech recognition model on 1,500 hours of curated clinical audio using the NVIDIA NeMo framework and 8x H100 GPUs, replacing its closed-source vendor ASR. The in-house Nemotron-based ASR stack cut end-to-end transcription latency from about 3.0 seconds to 0.7 seconds (75%+ improvement), reduced word error rate from 13% to 9.4%, and cut operating costs by 64% by eliminating per-minute API pricing, while giving Heidi full ownership of model weights for data sovereignty.
Was this helpful?
Your feedback helps us improve our use case database

