Project Year
Industry
A UAE-based software house aimed to develop an AI-powered healthcare solution for multiple hospitals. The product’s core functionality was to transcribe patient-doctor sessions, generate structured summaries, and store the data securely for future reference. The backend was developed using NestJS, while the frontend was built with Next.js.
The system leveraged Azure Transcription to process audio recordings and incorporated a form for doctors to input physical examination details. The combined data was then processed using ChatGPT to generate summaries and insights. The project evolved to support multi-tenancy, feature flagging, real-time monitoring, and HIPAA compliance.
The initial approach of processing audio chunks in real time led to memory leaks and high latency.
The system needed to support multiple hospitals while maintaining data segregation.
The client introduced an external system for user management, requiring a seamless migration without data loss
Patient data security had to meet strict healthcare standards.
Ensuring smooth system performance and quickly identifying issues was critical.
Solution Approach
Initially, audio chunks were sent from the frontend to the backend, which then forwarded them to the transcription service over WebSockets. However, this created performance bottlenecks. The solution:
Audio chunks were sent directly to the transcription service.
The transcription service processed audio in real time and stored the full audio in AWS S3.
Once the session ended, the complete transcription and audio URL were sent to the backend, improving efficiency.
To support multiple hospitals, multi-tenancy was introduced:
User authentication was migrated to the software house’s external system.
The user table in our database retained only the external system’s user ID, mapping it to existing records.
Only admins could create organizations, invite doctors, and manage permissions.
Custom redirect URLs were configured for each hospital.
To customize services for different hospitals:
Unleash (Self-Hosted) was integrated for feature flag management.
Enabled or disabled services based on hospital requirements.
Allowed seamless toggling of features without redeploying the system.
Datadog Integration: Monitored API requests, flagged errors, and provided real-time alerts for system failures.
Transaction Tracking: Logged server-side transactions to diagnose and resolve bottlenecks
SonarQube for Code Quality: Integrated into CI/CD pipelines to enforce high code quality and run unit tests.
HIPAA Compliance & Secure Data Handling
FHIR Standard for Patient Data: Patient information was stored securely using healthcare industry best practices.
PHI Data Encryption: Any Personally Identifiable Health Information (PHI) required by HIPAA was encrypted before storage.
Access Controls: Ensured that only authorized personnel could access sensitive patient records.
To enhance development efficiency and maintain high code quality, we leveraged GitHub Copilot along with Claude Sonnet 3.5 for AI-powered code generation. These tools helped:
Speed Up Development: Assisted in writing boilerplate code and optimizing functions.
Improve Code Consistency: Suggested best practices and efficient patterns.
Enhance Debugging & Refactoring: Provided intelligent recommendations for performance improvements.
Developers shared updates and discussed blockers.
Tasks were refined based on evolving requirements.
Fortnightly sprint planning with effort estimation.
Automating X-ray and CT scan analysis using AI models.
Exploring next-gen models to further reduce latency
Enhancing ChatGPT-generated reports with predictive healthcare trends.
This case study demonstrates how AI-driven automation, secure data handling, and scalable architecture can transform healthcare transcription and record-keeping, improving efficiency and compliance in medical institutions.
