JustSync.ai
Medical AI Platform • ML Engineering
Role: ML Engineering Intern
Duration: Mar 2024 – Aug 2024
The Problem
Healthcare professionals were spending hours manually searching through thousands of medical research papers and articles to find relevant information for patient care. The existing search tools were keyword-based and often missed contextually relevant research, leading to inefficient workflows and potentially outdated treatment approaches.
Challenge
The challenge was to create an intelligent medical information retrieval system that could understand context, not just keywords, while handling the complexity of medical terminology and maintaining accuracy. Additionally, the system needed to be deployed cost-effectively and integrated into both web and mobile platforms for accessibility.
Solution
I built a comprehensive medical AI platform consisting of three main components: (1) An intelligent data collection system that scraped and processed medical literature, (2) A contextual recommendation engine powered by fine-tuned language models and vector search, and (3) A cross-platform mobile application that made these insights accessible to healthcare professionals on the go.
Process
- Data Collection & Preprocessing (Month 1-2): Developed web scraping infrastructure to collect 2,500+ medical articles and research papers. Implemented data cleaning and preprocessing pipelines to prepare text for model training.
- Model Development (Month 3-4): Fine-tuned language models for medical domain understanding. Built and optimized the recommendation engine using LangChain and Gemini for contextual understanding.
- Vector Database & Search (Month 4-5): Implemented FAISS vector store for efficient similarity search. Optimized indexing strategies for cost-effective deployment on Google Cloud Run.
- Mobile App Development (Month 5-6): Developed React Native application integrating ML recommendations. Published to both App Store and Google Play Store with comprehensive testing.
Challenges Overcome
- Medical Text Complexity: Fine-tuned models specifically on medical literature and implemented domain-specific preprocessing to handle medical terminology accurately
- Cost-Effective Deployment: Optimized FAISS indexing and implemented efficient caching strategies to minimize cloud computing costs while maintaining performance
- Mobile App Store Approval: Ensured compliance with health app guidelines and implemented proper user privacy protections for both platforms
Impact & Results
- User Engagement: N/A (new platform) → 20% increase (Above industry average for medical apps)
- Data Coverage: 0 articles → 2,500+ articles indexed (Comprehensive medical knowledge base)
- Search Accuracy: Keyword-based search → Contextual AI recommendations (Semantic understanding implementation)
Key Achievements
- Fine-tuned ML models using 2,500+ medical articles and research papers for accurate data retrieval
- Developed and launched React Native app published on both App Store and Google Play Store
- Created keyword-based recommendation engine increasing user interaction by 20%
- Implemented cost-effective FAISS vector store deployment on Google Cloud Run
Tech Stack
React Native, LangChain, Google Gemini, FAISS, Google Cloud Run, Machine Learning, Python, Vector Databases, Web Scraping, Natural Language Processing