HealthForge

We built HealthForge to solve the problem of retaining insight from doctors appointments. When a patient leaves an appointment they forget a lot, according to the National Library of Medicine between 40-80% of information shared is forgotten upon leaving the doctors office. Our project retains insights generated through the use of voice to text transcription during the appointment itself, as well as AI summarization using the Claude API. We use a human in the loop framework that requires doctors to approve transcriptions, and AI summaries before allowing the patient to view it. Doctors can also add addendums to the AI insights if they believe important information was missed. The summaries are created through two calls, firstly to create the summary and then to evaluate it on a scale of 100; higher confidence indicates that a greater portion of the information in the AI report is inline with the transcription.

The primary actor in what can go wrong is AI hallucinations. This risk is mitigated by requiring a human in the loop with the doctor vetting what the user is able to see. By requiring approval before the patient can view the AI report allows for the data to have a lesser chance of being victim to AI hallucination. Another problem we expect is the sharing of sensitive information with Claude API services. To prevent this we would use a data anonymisation tool to prevent PII from being shared with Claude. This action ensures the privacy of the user which is a high priority in the medical industry.

HealthForge does not make decisions for you, medical decisions are still in the hands of your doctor. Our product merely stores conversations and insights between you and your healthcare professional.Through our sharing feature, you can even share insights with family members or get second opinions from other healthcare professionals in a secure and timely manner. The patient holds the power to revoke permissions from any family member to their personal medical information.