TEAM
Sunny Lee - Designer & Development
Danny Lee - Emergency Medicine Consultant & Quality Assurance

DURATION
3 weeks

TOOLS
Figma, Replit, Cursor, Claude, Chat GPT

PLATFORM
Next.js Web Application

 
 
 

OVERVIEW

In December 2024, my brother (an ER doctor) and I discussed how AI could help emergency physicians chart more efficiently. I was surprised to learn that many rely on trained medical scribes, costing around $3,000 per month. Hospitals typically covered this expense, but my brother recently learned this support will soon end. Going forward, he must either pay for a remote scribe himself, often based overseas, or use an AI scribe the hospital will provide at no cost.

 
 
 
 

DESIGN CONSIDERATIONS

To understand the competitive landscape, I analyzed platforms like Ambience, Commure, and Freed. Although I couldn't test these applications firsthand due to limited access, I reviewed their landing pages and feature lists. All three emphasized HIPAA compliance and mentioned Business Associate Agreements (BAAs) with their API partners.

 
 

Mock landing page designed in Figma, and analyzed by Vercel for quick front-end starter code.

 
 

v0

Before securing BAAs, I evaluated multiple transcription APIs, prioritizing cost and accuracy with medical terminology. The initial goal was to support English, with plans to gradually expand into other languages based on physician feedback over time.

 
 

My first engineering spike investigating transcription APIs.

 
 

MVP

On May 1, 2025, we launched Tricordr as a functional web application prototype to a select group of testers. What sets Tricordr apart is its ability to generate comprehensive documentation including HPI, MDM, ROS, physical exam notes, patient history, disposition plans, and diagnostic codes from a single intake recording. As a web based solution, it works seamlessly across all devices while allowing medical professionals to monitor recordings in real time. This transparency helps clinicians identify and address any recording or transcription gaps after patient encounters.

 
 
 

LESSONS LEARNED

  • All patient-related data is de-identified before being processed by the language model. The model does not receive or retain any Protected Health Information (PHI).

  • Under HIPAA guidelines, a Business Associate Agreement (BAA) is required for each API handling protected health information (PHI).

  • End-to-end encryption and an annual security plan are required for PHI, which must typically be retained for around seven years, dependent on state regulations and patient age.

 
 

CASE STUDIES

ENVISION

VIEW IN ROOM

MOBILE AR

SMART CAMPUS

RESPONSIVE WEB

BESPOKE SERVICES