HealthcareSaaS · Web application · AI/ML · API/backend · Infrastructure
AI medical scribe SaaS platform
We rescued and rebuilt an AI medical scribe for clinicians: it records encounters, transcribes them and generates editable, specialty-specific SOAP notes with ICD-10 codes. Alongside a React front end, subscription billing and video-meeting ingestion, we set up a secure AWS platform with encrypted storage, queued processing and automated API tests.
Problem
The client's previous developers had left the product undeployable, with no accessible git repository, shared root credentials and code that did not match production, while the client needed a reliable, secure product in time for investor and clinical demos.
Solution
The team recovered the source into AWS CodeCommit and redeployed it, then migrated the front end from Django templates to React, built the recording and transcription pipeline, engineered specialty-specific note-generation prompts on OpenAI, added Stripe billing and Zoom ingestion, and stood up dev and production AWS environments with encrypted storage, queued background processing, logging and single sign-on for engineers.
Key features
- Encounter recording with pause, resume and discard across sessions
- Transcription with PII redaction, later real-time streaming medical transcription
- SOAP note generation with specialty prompts (general practice, mental health, naturopathy, fertility)
- ICD-10 code extraction, editable notes and clinician e-signature
- Stripe subscription billing with a clinician onboarding flow
- Zoom cloud-recording ingestion via OAuth app and webhooks
- S3 artifact storage encrypted with KMS
- SQS-based background note processing with a Dockerized listener
- CloudWatch logging, Parameter Store configuration and IAM Identity Center SSO
- Automated URL-coverage API test suites and React unit tests
- EULA and privacy-policy acceptance tracking
- Figma designs across the whole product
Tech stack
- ACM
- AWS
- AWS Transcribe Medical
- AssemblyAI
- Bootstrap
- Celery
- Clerk
- CloudWatch
- CodeCommit
- Django
- Django REST Framework
- Docker
- EC2
- ECR
- ELB
- Figma
- GPT-3.5
- GPT-4
- IAM Identity Center
- KMS
- MUI
- Material UI
- NATS
- OpenAI
- Parameter Store
- PostgreSQL
- Python
- RDS
- React
- Redis
- Redux Toolkit
- Route53
- S3
- SQS
- Secrets Manager
- Stripe
- Vite
- WebSocket
- docker-compose
- nginx
- watchtower
Services behind this project
Read more
22SEPHealthcare software development: what to plan for
Healthcare software development means planning for patient privacy, HL7 and FHIR interoperability and audit trails from day one. Here is what to settle first.
Engineering6 min read
01SEPRescuing an inherited codebase: what we check in the first two weeks
A recurring kind of engagement: a product that already exists, a broken handover and a deadline that hasn't moved. Here's the checklist we run before we touch a line of code.
Engineering4 min read
04AUGShipping AI features into real products
Patterns that repeat across the AI products we've built: transcription pipelines, retrieval over your own data, a human kept in the loop, and controls on what the model costs and can see.
AI & Automation3 min read