Dispatch № 04 · scribeagent
ScribeAgent
AI medical transcription platform · HackGT
- Next.js
- FastAPI
- Python
- TypeScript
- WebSockets
A hackathon AI scribe built at HackGT that listens to a clinician's dictation and returns structured clinical data: a transcript, SOAP notes, billing codes, and prescriptions.
By the numbers
- Built a full stack medical transcription platform at HackGT (FastAPI and Next.js) that turns clinical dictation into structured SOAP notes, ICD-10 billing codes, and prescriptions.
- Streamed live transcription over WebSockets with a file upload fallback, taking raw audio to structured clinical data end to end.
The problem
Clinicians spend hours turning spoken notes into structured documentation. We wanted a tool that captures a dictation once and emits everything downstream systems actually need: a clean transcript, a SOAP note, ICD-10 billing codes, and a medication list.
Real time pipeline
A FastAPI backend handles two ingestion paths: live audio streamed over WebSockets for real time transcription, and a traditional file upload endpoint. Audio is transcribed and then post processed by an LLM layer into structured clinical fields.
Structured output
Beyond the raw transcript, the system generates SOAP notes, looks up ICD-10 billing codes, extracts structured prescriptions, and drafts lab orders, the parts that normally cost a clinician documentation time.
Stack
FastAPI and Python on the backend, Next.js and TypeScript on the front end, WebSockets for the live streaming path. Built under hackathon time pressure with team E = mc² + AI.