Helios

Solution

Helios worked with BCG on a generative-AI proof of concept for hospital discharge letters. The joint team first mapped more than 50 friction points in the existing workflow and then worked with Helios IT to acquire the patient data doctors review before writing a letter from the hospital information system. Because the project handled sensitive clinical data, BCG built the cardiology proof of concept in an on-the-premises environment. The implementation exported, pseudonymized, cleaned, and structured several thousand patient cases and more than 120,000 documents, then fine-tuned two open-source large language models on real patient data and historic discharge letters. The fine-tuned LLMs generated discharge-letter drafts that health care professionals validated, giving Helios a tested document-generation workflow and practical lessons for scaling GenAI in its broader digital strategy.

Data flow

Patient data that doctors would normally assemble from the hospital information system was exported, pseudonymized, cleaned, and structured alongside historic discharge letters. Those curated records were used to fine-tune open-source LLMs in an on-premises cardiology deployment. The models generated discharge-letter drafts, which health care professionals reviewed and validated before Helios planned broader rollout.