AI for teraMedik
This section of my internship involved a confidential project under a Non-Disclosure Agreement (NDA). The work focused on developing advanced data processing and machine learning techniques tailored to the company’s proprietary hospital information management system (HIMS).
Technical Focus
The project primarily focused on implementing scalable data pipelines and machine learning model architectures suitable for healthcare data environments. I also explored continuous model optimization workflows, API deployment standards, and strategies for adaptive learning in production systems.
Key Responsibilities and Achievements
While specific details and datasets cannot be disclosed, my contribution as a Machine Learning Programmer Intern included:
- Researched and compared multiple algorithms, including neural networks, to determine the most suitable approach for the project.
- Designed a comprehensive seven-step implementation plan covering data loading, preprocessing, model training, evaluation, and deployment.
- Built and validated dummy implementations using ~50,000 synthetic medical records to test code logic and algorithm performance.
- Processed and validated over 2.9 million real-world medical records for model development.
- Developed and optimized two health-related recommendation models (with and without neural networks) using PyTorch and MLX.
- Deployed final models as FastAPI REST APIs, resulting in four production-ready endpoints for real-time integration.
- Explored continuous learning frameworks (H2O.ai, Optuna) for potential future integration.
- Delivered weekly technical updates and comprehensive documentation to the project supervisor.
Reflection
This project deepened my understanding of real-world ML system development, especially in managing sensitive medical data and aligning research ideas with production constraints. It strengthened my ability to design clean, reusable, and deployment-ready ML codebases under tight data privacy regulations.
