Projects & Research
Welcome to my AI research portfolio! Here you’ll find detailed information about my research and development work in medical AI, natural language processing, computer vision, and machine learning.
Featured Research Projects
DSAL-Net: Dual-Stage Active Learning for Medical Segmentation
Research Focus: Weakly Supervised Learning, Medical Imaging, Active Learning
DSAL-Net introduces a novel dual-stage active learning framework specifically designed for breast ultrasound segmentation. This first-author research project tackles the critical challenge of limited labeled data in medical imaging through innovative CAM filtering and Mean Teacher consistency learning.
Key Innovations:
- Dual-stage architecture combining weak supervision with active learning
- HSV color space stabilization for robust pseudo-label generation
- Mean Teacher framework adapted for medical imaging consistency
- Performance: Achieved 68.25% IoU and 79.39% DSC on BUSI dataset
Research Status: First-author paper accepted at IEEE-BIBM 2025
Technologies: PyTorch, SAM, Computer Vision, Medical Imaging
Impact: Reduces annotation costs by up to 60% while maintaining clinical accuracy
MBTI-LLM: Personality-Controlled Text Generation
Research Focus: Large Language Models, Personality Psychology, Human-AI Interaction
Simulating human-like personality in LLMs using MBTI framework with sophisticated style control mechanisms.
Applications: Psychology studies, Human-AI interaction research, Personalized content generation
Technologies: Transformers, PyTorch, Natural Language Processing
Innovation: Systematic MBTI-based personality modeling for language models
Medfusion: Advanced Medical Image Generation
Research Focus: Generative AI, Medical Imaging, Synthetic Data
Medfusion transforms synthetic medical imaging through advanced VAEs and Diffusion Models. Specifically optimized for breast ultrasound synthesis, it generates realistic, tumor-inclusive outputs with precise distribution control for medical AI training.
Technical Highlights:
- Dual-Model Architecture: VAE embedder + conditional diffusion model
- Medical Optimization: Pathology-aware training for realistic anatomical structures
- Quality Assurance: 512×512 resolution with comprehensive evaluation metrics
- Clinical Validation: High acceptance rate from radiologist evaluation
Technologies: Diffusion Models, VAE, PyTorch, Medical Image Processing
Period: September 2024 - December 2024
HandSense-360: Real-time Gesture Recognition
Research Focus: Computer Vision, Healthcare Applications
Advanced computer vision system for healthcare and control applications.
Technical Capabilities:
- Real-time gesture tracking with MediaPipe integration
- Precision metrics and seamless external system integration
- Healthcare innovation applications for rehabilitation
- Advanced control systems for accessibility
Applications: Healthcare rehabilitation, Accessibility tools, Human-computer interaction
AI Applications & Tools
DeepTeacher: AI-Powered Learning Assistant
Local automatic learning assistance tool combining llava-phi3:3.8b and DeepSeek-R1 models with real-time screen capture, image understanding, and intelligent feedback.
Core Features:
- Real-time screen capture and image understanding capabilities
- Intelligent language feedback system
- Local deployment ensuring privacy protection
- Multi-modal learning support
Technology Stack: llava-phi3:3.8b, DeepSeek-R1, Computer Vision, Natural Language Processing
DeepPlaylist: Intelligent Music Classification
Automatic playlist classification system based on DeepSeek and QQ Music API with customizable classification rules.
Key Features:
- Default classification by language and music mood
- Customizable classification methods and prompts
- Intelligent music content analysis
- API integration and automated processing
Applications: Music recommendation, Content management, Personalized playlists
MagicSub: AI-Powered Subtitle Translation
Advanced subtitle translation system leveraging state-of-the-art language models for accurate content localization across multiple languages.
Core Capabilities:
- Multi-language subtitle translation with context awareness
- Batch processing for large video libraries
- Customizable translation models and quality settings
- Support for multiple subtitle formats (SRT, VTT, ASS)
Technical Features:
- Integration with leading translation APIs
- Advanced timing synchronization algorithms
- Quality assurance through back-translation validation
- User-friendly GUI with drag-and-drop functionality
Applications: Content creators, Educational platforms, Media companies, International distribution
Machine Learning & Data Science
Diabetes Prediction: Multi-Algorithm Comparison Study
Comprehensive machine learning study comparing 10+ algorithms for diabetes prediction with extensive feature engineering and model evaluation.
Algorithms Implemented:
- Support Vector Machines (SVM) with multiple kernels
- Stochastic Gradient Descent with various optimizers
- Ensemble methods (Random Forest, XGBoost)
- Neural networks (MLPClassifier with multiple architectures)
- Traditional methods (Logistic Regression, Naive Bayes, KNN)
Research Contributions:
- Systematic comparison of algorithm performance on medical data
- Feature importance analysis and selection strategies
- Cross-validation and hyperparameter optimization
- Statistical significance testing of model differences
Performance Metrics: Accuracy, Precision, Recall, F1-Score, AUC-ROC analysis across all models
Technologies: Python, Scikit-learn, XGBoost, Pandas, NumPy, Matplotlib
Research Impact & Open Source Philosophy
Technical Expertise
- Programming Languages: Python, Java, C#, MATLAB, SQL
- AI/ML Frameworks: PyTorch, TensorFlow, Transformers, Diffusers
- Specializations: Medical AI, Computer Vision, NLP, Generative Models
- Research Areas: Weakly Supervised Learning, Active Learning, Personality AI
Open Source Contribution
- Total Repositories: 12+ AI-focused projects
- Research Papers: 1 first-author submission accepted
- Technologies: Medical AI, LLMs, Computer vision, Machine learning
- Impact: Projects used in academic research and real-world AI applications
Collaboration & Contact
Research Interests: Medical AI, Personality Modeling, Generative Models, Computer Vision
Open to: Research collaborations, Academic partnerships, Industry applications
Contact: Get in touch for project discussions or collaboration opportunities
All projects are open source and available under permissive licenses. Star the repositories you find interesting!