GitHub 🐱

Slack 📡

I am a data science engineer with expertise in machine learning, computer vision, and collaborative research. My background includes study in internet and network technologies, with a proven track record in developing AI-driven solutions for complex real-world problems. I am passionate about leveraging data and technology to create meaningful, clinically-validated applications that improve lives.

Languages & Frameworks:

• Python (TensorFlow, Keras, PyTorch, Pandas, NumPy)
• Data Analysis & Visualization (Matplotlib, Seaborn)
• Computer Vision & Image Processing (OpenCV, Pillow)
• Natural Language Processing techniques

Core Competencies:

• Deep Learning & Model Optimization
• Data Science & Statistical Analysis
• Computer Vision & Image Classification
• Clinical Data Validation
• Collaborative Research & Development

Technical Interests:

• Exploring diverse programming languages and paradigms
• Learning through practical application and experimentation
• Building robust, well-researched solutions
• Cross-functional team collaboration

Repository

Overview: A research-driven AI system using computer vision to automatically identify food items and estimate carbohydrate content for Type 1 Diabetes management. This clinically-validated solution aims to reduce the cognitive burden of manual carbohydrate counting, a critical component of insulin dosing decisions.

Key Achievements:

• Implemented EfficientNetB0 deep learning model with 91.9MB footprint
• Created gold standard dataset with 5,000+ validated, clinically-reviewed images
• Optimized for consumer-grade GPU deployment (RTX 3050, 3.2GB VRAM)
• Achieved sub-100ms inference speed with mixed precision (FP16) training
• Implemented clinical safety auditing with carbohydrate bucketing validation

Technical Architecture:

• Framework: TensorFlow 2.15+ / Keras
• Optimization: Mixed Precision Training, Memory Limiting
• Computer Vision: OpenCV, Pillow
• Interface: Real-time camera integration with Tkinter GUI
• Hardware Target: NVIDIA RTX 3050/3000 series (3.2GB VRAM)

Project Goals:

1. Build clinically-validated gold standard dataset
2. Optimize for consumer hardware deployment
3. Ensure clinical safety through carbohydrate bucketing
4. Achieve high classification accuracy and reliability
5. Deploy user-friendly real-time applications

Repository

Overview: A collaborative data science research project analyzing conversational failures in mental health chatbots. Conducted as part of the MIT Emerging Talent Collaborative Data Science Program (CDSP), this research explores the effectiveness and limitations of AI-driven versus human support in digital mental health applications across diverse cultural contexts.

Research Focus:

• Program: MIT Emerging Talent CDSP (Track 6)
• Domain: Digital Mental Health & Emotional Support Technologies
• Timeline: May – August 2025
• Team: DataCure – A cross-cultural research collaboration

Research Question:

• What are the most prevalent themes of user-reported conversational failure in leading mental health chatbots, and what do these themes reveal about the gap between user expectations for emotional support and current algorithmic capabilities?

Key Objectives:

1. Systematically identify and categorize common user-reported chatbot failures
2. Quantify failure prevalence using public app store and forum data
3. Compare failure patterns between conversational AIs and baseline wellness apps
4. Generate actionable insights for developing more empathetic digital care tools

Collaborative Team:

Aziz Azizi, Huda Alamassi, Malak Battatt, Sadam Husen Ali, Chrismy Augustin, F. Ismail SAHIN, Ayham Hasan,

Technical Stack:

• Python (Pandas, NumPy, Scikit-learn)
• Jupyter Notebooks for iterative analysis
• Natural Language Processing
• Data visualization (Matplotlib, Seaborn)