Inspiration
Healthcare providers spend an enormous amount of time on paperwork, especially when it comes to filling out an insurer's claim and prior authorization forms. This administrative burden delays treatments and takes away from healthcare professionals' time that could be spent with patients. Our project, MedRelay, hopes to reduce that administrative friction.
What it does
MedRelay listens to natural doctor-patient conversations, transcribes them, and then:
- Extracts key information: diagnoses, procedures, prescriptions, insurance info.
- Identifies the medical outcomes: new prescriptions, referrals, tests ordered.
- Maps that data directly into standardized forms.
The result is a draft form that a provider can quickly review and submit — saving valuable time while reducing human error. Most importantly, MedRelay emphasizes the power of AI in human-in-the-loop systems. Our project doesn't replace providers; it speeds them up.
How we built it
Speech-to-Text: Using transcription models to capture doctor–patient dialogue in real time.
Entity Recognition: We applied natural language processing to extract and link structured data (symptoms, ICD-10 codes, CPT codes, insurance coverage) to persistent patient/provider information.
Form Mapping Engine: That structured data is automatically inserted into the appropriate fields of CMS-1500 and CMN forms.
Challenges we ran into
Medical conversations often contain ambiguous or colloquial phrasing (“my knee locks up sometimes”) that needed to be mapped to precise codes.
The variety of forms and requirements (CMS-1500, CMN, prior authorization paperwork) made field-mapping tricky.
Additionally, we had to be extra careful to store and encrypt patient + provider data responsibly, especially in the context of LLMs.
Accomplishments that we're proud of
Built a functional pipeline from voice → transcription → structured data → form completion, backed up by a clean UI and workflow.
Designed a system that can scale to multiple forms and use cases in the future.
Log in or sign up for Devpost to join the conversation.