top of page
Projects
Showcasing comprehensive Machine Learning, Data Science, and AI portfolio implementations featuring predictive modeling, computer vision, NLP, and scalable software development solutions.
1️⃣ AI-Powered Multimodal Chronic Disease Risk Screener
Problem
- Traditional chronic disease screening requires invasive tests, expensive equipment, and specialized medical facilities, limiting accessibility and early detection in underserved populations
- Current screening methods often detect diseases at advanced stages when treatment options are limited and outcomes are poorer
Solution
- Developed a production-ready, non-invasive screening system combining retinal imaging, nailbed images, and demographic/lifestyle data to assess diabetes, hypertension, and diabetic retinopathy risk
- Created an accessible, cost-effective screening solution that can be deployed in primary care settings and remote locations
System Design & Implementation
- Implemented CNN retinal models (primary + robustness model) for diabetic retinopathy detection with transfer learning optimization
- Built EfficientNet-based nailbed classifier for diabetes and hypertension risk assessment using computer vision techniques
- Developed gradient-boosted tabular model for processing demographic and lifestyle health factors
- Designed calibrated fusion layer aggregating multimodal predictions into patient-level risk scores with explainability features
Impact & Results
- Achieved ROC-AUC 0.999 for diabetic retinopathy detection, exceeding clinical benchmark standards
- Delivered 91% nailbed classification accuracy for diabetes and hypertension screening
- Obtained 0.91 ROC-AUC for hypertension prediction, demonstrating strong predictive capability
- Provided interpretable outputs including feature importance, calibration curves, and risk breakdowns for clinical decision support
- Reduced screening time from hours to minutes while maintaining clinical-grade accuracy
Tech Stack
Python, TensorFlow, Keras, EfficientNet, XGBoost, Scikit-learn, OpenCV, Pillow, Albumentations, FastAPI, Uvicorn, React.js, Next.js, Pandas, NumPy, Docker, AWS, GCP
2️⃣ AI-Based Multi-Organ Segmentation in Medical Imaging
Problem
- Manual medical image segmentation for liver, pancreas, and tumors is time-intensive and error-prone
- Existing automated tools lack clinical-grade accuracy for multi-organ tasks
- Need for scalable solution that radiologists can trust and integrate into workflows
Solution
- Developed AI-powered segmentation system for automated liver, pancreas, and tumor detection from abdominal CT scans
- Combined deep learning research with clinical usability requirements
- Created production-ready pipeline for real-world medical imaging deployment
System Design & Implementation
- Designed UNet–ResNet34 hybrid architecture combining representation learning with spatial accuracy
- Implemented pixel-level class balancing to handle extreme tumor sparsity (0.3% of pixels)
- Built patch-based, mixed-precision training pipeline for scalable, memory-efficient learning
- Applied progressive multi-stage training to stabilize convergence across heterogeneous organs
Impact & Results
- 88.2% validation Dice and 86.7% test Dice across multi-organ segmentation tasks
- Scaled training to 69,508 CT slices with consistent performance
- Reduced annotation time from hours to under one minute per scan
- Demonstrated clear path from research prototype to deployable system
Tech Stack
Python, PyTorch, UNet, ResNet34, CUDA, Albumentations, NumPy, OpenCV
3️⃣ Smart Rerouting System for Metro Passengers
Problem
- Urban commuters face unpredictable delays due to dynamic transportation disruptions
- Existing metro apps lack real-time integration with traffic and weather conditions
- Manual route planning becomes inefficient during rush hours and unexpected service changes
Solution
- AI-powered route optimization system combining multiple real-time data sources
- Intelligent recommendation engine that adapts to current conditions
- Full-stack dashboard for seamless user experience across devices
System Design & Implementation
- Developed microservices architecture for scalable real-time data processing
- Integrated multiple APIs: metro scheduling, traffic conditions, weather forecasts
- Implemented intelligent caching and fallback mechanisms for 99.9% uptime
- Built responsive dashboard with real-time updates and push notifications
- Deployed machine learning algorithms for predictive route optimization
Impact & Results
- Reduced average commuter delays by 18% during peak hours
- Achieved 99.9% system uptime with intelligent failover mechanisms
- Processed over 10,000 route requests per day with sub-second response times
- Led cross-functional team of 4 developers through agile development process
- Delivered project 2 weeks ahead of schedule within budget constraints
Tech Stack
Python, Flask, React, Node.js, PostgreSQL, Redis, REST APIs, Docker, AWS
4️⃣ Intelligent Urban Traffic Management System
Computer Vision | Applied Machine Learning | Smart City Research
Problem
- Rapid urbanization has made real-time traffic density estimation critical for congestion control and intelligent transportation systems. Traditional monitoring approaches lack scalability, adaptability, and accuracy across diverse traffic conditions.
Solution
- Designed an AI-powered traffic management system that automatically detects vehicles and classifies traffic density patterns using pretrained deep learning models, enabling data-driven, real-time traffic insights.
System Design & Methodology
- Implemented YOLOv8x for high-precision, real-time vehicle detection
- Utilized MobileNet with transfer learning for efficient traffic density classification
- Built an automated image analysis pipeline covering preprocessing, detection, and classification
- Optimized for deployment in urban traffic surveillance environments
- Evaluated across varied traffic scenarios to ensure robustness and generalization
Results & Impact
- Achieved 94.75% test accuracy in traffic density classification
- Enabled reliable real-time traffic analytics for congestion management
- Reduced manual intervention through full automation
- Demonstrated applicability to smart cities and intelligent transportation systems (ITS)
Tech Stack
Python, YOLOv8, MobileNet, OpenCV, CNNs, Transfer Learning, Computer Vision, Data Analytics
Get in Touch
Open to full-time roles and internships in AI, Machine Learning, and Software Engineering. Recruiters and hiring managers can reach me directly at niyatikapadia111@gmail.com or via LinkedIn
bottom of page