AWS re:Invent 2025-How Heidi Health is leveraging GenAI to transform the global healthcare industry

Summary of AWS re:Invent 2025 Presentation: "How Heidi Health is Leveraging GenAI to Transform the Global Healthcare Industry"

Introduction to Heidi Health

• Heidi Health is a startup aiming to solve the problem of clinicians being bogged down by administrative tasks, allowing them to focus more on patient care.
• Heidi's mission is to "double the world's healthcare capacity by providing an AI care partner."
• Heidi's founder, Tom Kelly, started by building a chatbot tool called Oscar to help medical students with clinical examinations using early transformer models.
• Heidi then expanded into a full healthcare platform to support doctors and enhance patient care, until the emergence of generative AI (GenAI) enabled them to focus on clinical note generation.

Heidi's GenAI-powered Clinical Note Generation

• Heidi's ambient AI scribe, Heidi, can automatically generate clinical notes during doctor-patient consultations without any further action needed.
• Heidi can also provide doctors with suggested follow-up tasks and clinical research based on the consultation.
• This allows doctors to focus on the patient interaction while Heidi handles the administrative burden of note-taking and documentation.

Scaling Challenges and Solutions

Building Confidence in AI

• Heidi faced the challenge of validating the clinical accuracy and empathy of their GenAI-generated notes at scale.
• They started by providing doctors with Jupyter notebooks to experiment and test the AI, but this was not scalable.
• Heidi then developed internal tooling to enable clinicians to evaluate flag sessions, understand the AI's context, and use the AI as a judge to evaluate at scale.
• This feedback loop helped improve the model prompts and ensure medical safety.

Scaling Globally

• Heidi encountered four key challenges when expanding globally:
  1. Data sovereignty: Ensuring data is stored and processed within specific regional boundaries.
  2. Model availability: Adapting to the availability of compliant GenAI models in different regions.
  3. Medical reality: Adapting the consultation flow and medical terminology to local practices.
  4. Regulatory compliance: Navigating the shifting regulatory landscape across different regions.
• To address these challenges, Heidi:
  • Standardized their infrastructure using Infrastructure as Code for plug-and-play deployment.
  • Utilized a hybrid approach, using pre-existing compliant models (e.g., Amazon Bedrock) for speed, and self-hosted models on EKS for long-term control.
  • Hired clinician ambassadors in each region to ensure Heidi can speak the local medical dialect and understand local practice patterns.
  • Established a dedicated internal legal and compliance team to manage the shifting regulatory landscape.

Key Takeaways

1. Focus on solving one painful problem perfectly, rather than trying to "boil the ocean."
2. Treat subject matter experts (doctors and clinicians) as core assets, not just testers.
3. Build a flexible architecture from the start to enable expansion and adaptation to changing requirements.

Technical Details

• Heidi is used by over 370,000 clinicians, handling 10 million consults per month.
• Heidi is the #1 AI scribe by adoption in Canada.
• Heidi is backed by well-known investors.
• Heidi utilizes Infrastructure as Code, EKS, and a hybrid approach to GenAI models to enable global scalability.

Business Impact

• Heidi's AI-powered clinical note generation frees up doctors to focus on patient care, improving the overall patient experience.
• By doubling healthcare capacity, Heidi aims to have a significant impact on the global healthcare industry.
• Heidi's solutions have been particularly valuable during the COVID-19 pandemic, when clinicians were under increased pressure.

Examples

• Heidi can automatically generate clinical notes during doctor-patient consultations, allowing doctors to focus on the patient interaction.
• Heidi can provide doctors with suggested follow-up tasks and clinical research based on the consultation.
• Heidi has adapted its solutions to the unique requirements of different regions, such as data sovereignty, model availability, and medical terminology.