Healthcare solution brings Llama-powered clinical reasoning to patient care and record management
At a glance
Industry
HealthcareUse case
Streamlining patient assessments and automating documentation with AI-powered clinical reasoningGoal
Reduce administrative tasks and improve patient careLlama versions
Llama 3.1 8B, 70B and 405B; Llama 3.2 3BDeployment
Dedicated infrastructure on Oracle Cloud
Reimagining healthcare with clinical reasoning AI
Sofya works with cutting-edge healthcare institutions and researchers to push the boundaries of healthcare innovation and bioinformatics. Built on a solid research foundation, Sofya continues to publish academic papers and open source its models and datasets as it commercializes its clinical reasoning technologies.Help healthcare providers focus on patients instead of paperwork
Digital healthcare administration is overwhelming providers. Physicians spend 2.5 to 8.4 hours documenting care in electronic health record platforms for every eight hours of scheduled patient time. Across specialties, the average is 3.4 hours.Sofya saw an opportunity to give time back to providers by streamlining healthcare record management and providing AI-powered assistants that could enhance clinical reasoning during patient consultations.
Use Llama models to rationalize healthcare documentation, improve bedside care
Sofya developers fine-tuned a series of Llama models to create a multi-agent solution that helps structure healthcare data, applies clinical reasoning and assists providers during patient consultations. The Sofya clinical assistant guides providers through taking patient histories, flags concerns, offers diagnoses and creates personalized care plans.The team chose Llama models for their adaptability and capacity to produce high-quality clinical documentation. Open-source Llama freed Sofya developers to customize models and deploy them on their own infrastructure for maximum data control and security, a key requirement for the healthcare industry.
Fine-tune expert models for each role and application
Developers generated high-quality synthetic training data for the Sofya solution using Llama 3.1 405B and a combination of knowledge distillation and self-reflection prompts. To deliver clinical assistance from desktop to mobile, the solution uses trained and fine-tuned Llama 3.1 70B and 8B and Llama 3.2 3B models to balance latency, performance and computing resources.Llama models power the heart of the Sofya solution — a clinical reasoning system with humans in the loop that clarifies terms and structures healthcare information into unified, medical protocols and decision-support standards. These standards serve as programmatic rules that human caregivers, automated workflows and other applications can follow.
Time savings for providers, lower costs for organizations
Sofya customers report providers are spending less time on documentation and that overall costs have decreased. Healthcare providers report better, more efficient workflows and, most importantly, improved patient care.- 30% faster throughput in tokens per second vs. the previous model
- 90% customer satisfaction (CSAT) score
- 33% Healthcare providers report better workflow efficiency and patient care outcomes
